Georgia State University. Georgia State University. Anna Agapova

Transcription

1 Georgia Sae Universiy Georgia Sae Universiy Finance Disseraions Deparmen of Finance Cross-Secional Differences beween Topic 1: Money Marke Muual Funds and heir Role in he Muual Fund Families. Topic 2: Innovaions in Financial Producs. Convenional Muual Funds versus Exchange Traded Funds. Anna Agapova Follow his and addiional works a: hp://scholarworks.gsu.edu/finance_diss Recommended Ciaion Agapova, Anna, "Cross-Secional Differences beween Topic 1: Money Marke Muual Funds and heir Role in he Muual Fund Families. Topic 2: Innovaions in Financial Producs. Convenional Muual Funds versus Exchange Traded Funds.." Disseraion, Georgia Sae Universiy, hp://scholarworks.gsu.edu/finance_diss/10 This Disseraion is brough o you for free and open access by he Deparmen of Finance a Georgia Sae Universiy. I has been acceped for inclusion in Finance Disseraions by an auhorized adminisraor of Georgia Sae Universiy. For more informaion, please conac scholarworks@gsu.edu.

2 Permission o Borrow In presening his disseraion as a parial fulfillmen of he requiremens for an advanced degree from Georgia Sae Universiy, I agree ha he Library of he Universiy shall make i available for inspecion and circulaion in accordance wih is regulaions governing maerials of his ype. I agree ha permission o quoe from, or o publish his disseraion may be graned by he auhor or, in his/her absence, he professor under whose direcion i was wrien or, in his absence, by he Dean of he Robinson College of Business. Such quoing, copying, or publishing mus be solely for scholarly purposes and does no involve poenial financial gain. I is undersood ha any copying from or publicaion of his disseraion which involves poenial gain will no be allowed wihou wrien permission of he auhor. Signaure of auhor

3 Noice o Borrowers All disseraions deposied in he Georgia Sae Universiy Library mus be used only in accordance wih he sipulaions prescribed by he auhor in he preceding saemen. The auhor of his disseraion is: Anna Agapova Deparmen of Finance J. Mack Robinson College of Business Georgia Sae Universiy P.O. Box 3991 Alana, GA The direcor of his disseraion is: Jason T. Greene Associae Professor Deparmen of Finance J. Mack Robinson College of Business Georgia Sae Universiy 35 Broad Sree, Suie 1243 Alana, GA Users of his disseraion no regularly enrolled as sudens a Georgia Sae Universiy are required o aes accepance of he preceding sipulaions by signing below. Libraries borrowing his disseraion for he use of heir parons are required o see ha each user records here he informaion requesed. Name of User Address Dae 2

4 TWO ESSAYS ON MUTUAL FUNDS BY ANNA AGAPOVA A Disseraion Submied in Parial Fulfillmen of he Requiremens for he Degree of Docor of Philosophy in he Robinson College of Business of Georgia Sae Universiy GEORGIA STATE UNIVERSITY ROBINSON COLLEGE OF BUSINESS

5 Copyrigh by Anna Agapova

6 ACCEPTANCE This disseraion was prepared under he direcion of he candidae s Disseraion Commiee. I has been approved and acceped by all members of ha commiee, and i has been acceped in parial fulfillmen of he requiremens for he degree of Docor in Philosophy in Business Adminisraion in he Robinson College of Business of Georgia Sae Universiy. Disseraion Commiee: Jason T. Greene, Ph.D. Commiee Chair Vikas Agarwal, Ph.D. Commiee Member Gerald D. Gay, Ph.D. Commiee Member Conrad S. Ciccoello, Ph.D. Commiee Member Dean Robinson College of Business 5

7 TABLE OF CONTENTS ESSAY ONE..10 CROSS-SECTIONAL DIFFERENCES BETWEEN MONEY MARKET MUTUAL FUNDS AND THEIR ROLE IN MUTUAL FUND FAMILIES Inroducion Cross-secional examinaion of money marke muual funds Facors influencing MMMF reurns Family facors Cash managemen Loads and cash cener Risk managemen Clienele Family effecs Daa descripion and empirical resuls Conclusion References..43 ESSAY TWO.56 INNOVATIONS IN FINANCIAL PRODUCTS: CONVENTIONAL MUTUAL FUNDS VERSUS EXCHANGE TRADED FUNDS Inroducion Lieraure review Comparison of convenional and exchange raded index funds Hisory of ETFs Concepual differences beween ETFs and convenional open-end index funds Hypoheses formulaion Subsiuion effec Clienele effec Daa sources and descripive saisics Empirical analysis Performance Subsiuion effec Clienele effec Tax clienele Insiuional clienele Summary of resuls Conclusion Appendix References..87 Via

8 LIST OF TABLES AND FIGURES Table 1.1. Descripive Saisics of he MMMFs...45 Table 1.2. Number of MMMFs across Families 46 Table 1.3. Univariae Analysis of he Muual Fund Familys Risk, Reurn, Expenses, and Toal Ne Asses..47 Table 1.4. Descripive Saisics of Families wih MMMFs..48 Table 1.5. MMMFs Reurn Composiion.49 Table 1.6. MMMFs as a Cash Cener of a Fund Family 50 Table 1.7. Deerminans of MMMFs Expenses 51 Table1. 8. Effec of Family Loads on he MMMFs Characerisics.52 Table 1.9. Effec of he Family s Risk on MMMFs Risk and Mauriy...53 Table Effec of he Family Fund Invesmen Objecives on he Number and Invesmen Objecives of MMMFs 55 Table 2.1. Asses in Equiy Muual Funds and Exchange Traded Funds, Table 2.2. Univariae Analysis of Effeciveness and Tracking Error: Convenional versus Exchange Traded Index Funds...91 Table 2.3 Descripive saisics of ETFs and Convenional Index Funds Grouped by Index..94 Table 2.4. Subsiuion Effec: Exchange Traded and Convenional Index Funds Aggregae Flows 95 Table 2.5. Tax Clienele.96 Table 2.6. Insiuional Clienele 98 Figure Figure

9 ABSTRACT TWO ESSAYS ON MUTUAL FUNDS BY ANNA AGAPOVA April 16, 2007 Commiee Chair: Major Deparmen: Dr. Jason T. Greene Finance The firs essay examines cross-secional differences beween money marke muual funds (MMMFs), in he conex of he sponsoring fund family. While exan sudies have shown ha fund family characerisics impac he managemen of open-end equiy muual funds, resuls of his sudy s analysis find ha fund family characerisics also affec he managemen of MMMF asses, conribuing o differences in he mauriy of he fund s holdings, expenses, and realized reurns. I find ha an MMMF is no simply a ransiional accoun wih a shor-erm low-risk invesmen objecive, bu raher, a criical role player wihin he fund family. Differences in mauriy, yield, and expenses in MMMFs can be explained by family-specific characerisics, including diversificaion and cash managemen sraegies a he family level. The second essay examines implicaions of subsiuabiliy of wo similar financial asses: convenional index muual funds and exchange raded funds (ETFs). I seek o explain he coexisence of hese fund ypes, since boh offer a claim on he same underlying index reurn process, bu have differen organizaional srucures. This sudy compares convenional open-end index funds wih mached ETFs on various underlying indexes. Aggregae flows are used o deec subsiuion and clienele effecs. I show ha convenional funds and ETFs are subsiues, while ETFs have smaller racking errors and lower fund expenses. However, I find ha hese fund ypes are no perfec subsiues, and heir coexisence can be explained by a clienele effec ha segregaes hem ino differen marke niches. 8

10 ESSAY ONE CROSS-SECTIONAL DIFFERENCES BETWEEN MONEY MARKET MUTUAL FUNDS AND THEIR ROLE IN MUTUAL FUND FAMILIES 1. Inroducion Money marke muual funds (MMMFs) have exised for more han hree decades, wih he firs such funds being inroduced in By 1984, 305 MMMFs exised, oaling nearly $270 billion in asses. Over he las wo decades, MMMFs have grown o include more han $2 rillion in asses across 993 funds, and comprise approximaely 25% of U.S. open-end muual fund asses by 2005, according o he Invesmen Company Insiue (ICI). Though MMMFs are second only o equiy funds in erms of dollar value of asses in he muual fund indusry, he majoriy of exising lieraure concenraes on equiy funds. Many sudies of open-end equiy funds examine how hese funds differ in he cross-secion, and how hese funds are affeced by he characerisics of he sponsoring family. In conras, here are currenly only a few sudies ha address he cross-secional differences among money marke funds, and no sudies relaing MMMF fund family characerisics. However, numerous researches focus on he skill of equiy fund managers. 1 Alhough here is exensive lieraure on he cross-secional differences in performance and fund characerisics among equiy muual funds ha examines managemen effeciveness in efficien markes, here are virually no sudies ha address hese issues wih regard o muual funds ha inves in fixed income securiies, specifically, MMMFs. As exemplified in Table 1, here are subsanial differences in weighed average mauriy, 1 See, for example, Elon e al (1993), Carhar (1997), Chevalier and Ellison (1997), among ohers. 9

11 yields and expenses across MMMFs; however, relaively lile research has been done o explain hese cross-secional differences. This sudy exends he coverage of issues relaed o MMMFs. In his sudy, I apply fund family characerisics similar o hose ha have been used o explain differences among equiy funds, o explain cross-secional differences beween MMMFs. I propose ha money marke muual funds are no jus ransiional accouns wih shor-erm low-risk invesmen objecives, bu migh play an imporan role wihin he fund family. As he muual fund family has is own objecive funcion o maximize, i may use MMMFs o improve performance of he family in oal. Specifically, I examine he exen o which MMMFs are used wihin fund families for he purposes of cash and risk managemen. I also examine he impac of clienele effecs on he characerisics of MMMFs. To sudy he performance-flow relaion of MMMFs and how i affecs decisions o waive fees, Chrisoffersen (2001) employs Sirri and Tufano s (1998) mehodology of esimaing a piecewise-linear fund flow funcion and finds ha beer performers arac more flows, hough poor performers do no experience ouflows. Furher, Chrisoffersen examines he decision o waive fees and finds ha variaion in fee waivers is significan and relaes o he relaive performance of he MMMF. In his sudy, I examine wheher MMMFs play he role of a family s inernal cash cener in which oher funds in a family may perform heir liquidiy ransacions using MMMFs. By using such cash managemen sraegies, fund families can save on ransacion coss a he family level by avoiding ransacions wih exernal eniies when possible. Invesors can also use MMMFs similarly, as hey have an opion of free asse 10

12 ransfers wihin he family, wihou paying load fees. Thus, if family funds have high loads, hen invesors will likely prefer o say in he family and use is MMMFs o emporarily park heir cash before invesing i elsewhere, versus using sources ouside he family o mee ha need. In his sudy, I search for evidence of such invesor and fund family behavior and examine is exen, based on fund family characerisics. I also sudy he degree o which a family s risk may be inernally managed using MMMFs, by adjusing he mauriies and risk of he funds underlying securiies, and hereby adjusing he risk of MMMFs. I expec ha if fund families pursue such risk conrol sraegies, hen more concenraed families wih less diversified porfolios and riskier funds wih regard o oher invesmen objecives will have less risky MMMFs, holding shorer mauriy securiies. This sudy explores he evidence of fund family risk conrol in conjuncion wih is MMMFs, and suggess furher explanaions for fund behavior based on family characerisics. Cerain cross secional differences beween MMMFs wihin a family can also resul from a clienele effec, e.g. a family wih a larger variey of funds and invesmen objecives will more likely offer MMMFs wih various characerisics o capure he heerogeneiy of he family s invesors. Massa (2003) finds evidence of family-driven heerogeneiy among funds, and shows ha families acively exploi i. Resuls demonsrae ha MMMF reurns are influenced by risk and mauriy of he MMMFs porfolios, fund expenses, size, and macroeconomic facors. I find evidence ha fund families and heir invesors use MMMFs as a cash cener, and ha fund families use MMMFs o inernally manage risk a he family level. The observed diversiy in he number and ypes of MMMFs in families can be explained by differences in he number 11

13 of oher ype of funds offered wihin he complex, and by he variey of invesmen objecives for non-mmmfs in he fund family. Thus, differences in fund characerisics, such as mauriy, yield, and expenses across MMMFs can be explained by family-specific characerisics, including cash managemen and diversificaion sraegies a he family level. Applicaion of hese sraegies can reduce operaing coss and improve overall performance a he family level. The remainder of his paper is organized as follows: Secion 2 develops he hypoheses and mehodology for examining MMMFs. I provide daa descripion and empirical analyses in Secion 3, while Secion 4 concludes he paper and offers research implicaions. 2. Cross-secional examinaion of money marke muual funds MMMFs offer invesors a relaively homogeneous produc shor-erm deb securiies wih relaively low risk. Choice of securiies ha can be used in money marke porfolios is limied by regulaions. Rule 2a-7 under he Invesmen Company Ac of 1940 specifies ha money marke funds may no acquire insrumens wih remaining mauriy of greaer han 397 days or will no mainain a dollar-weighed average porfolio mauriy ha exceeds 90 days. The rule also specifies porfolio qualiy and diversificaion ha money marke funds are o mainain, e.g. funds mus limi invesmens o securiies wih minimal credi risk, and inves no more han five percen of he fund s oal ne asses in second ier securiies, or hose of a single issuer. As he producs are very similar, we would expec homogeneous characerisics among MMMFs, i.e. weighed average mauriy, yields and expense raios should no vary much. Therefore, one would 12

14 no expec widely differen sraegies and unique asse composiions across various money marke funds. Early research on MMMFs focuses on he porfolio manager s abiliy o predic ineres rae flucuaions by examining he associaion beween he porfolio s average mauriy and ineres rae changes. Ferri and Oberhelman (1981), and Packer and Pencek (1990) analyze aggregae daa for changes in MMMFs average mauriy and subsequen changes in CD raes, finding evidence ha managers, as a group, show some abiliy o predic fuure money marke yields. Domian (1992), and Seyfried and Packer (2001) sudy he causaliy of he mauriy-marke yield relaionship of MMMFs by uilizing Granger-causaliy ess and find ha a relaionship exiss in he opposie direcion from ha suggesed by previous sudies. However, hey sill find ha managers have he abiliy o predic changes in shor-erm ineres raes. While earlier sudies examine ime series variaions of MMMFs mauriies, I focus on cross-secional variaions beween MMMF characerisics. Consisen wih earlier sudies conclusions ha find ha MMMFs are acively managed, I conclude ha MMMFs do vary significanly in imporan characerisics, including mauriy, reurn, and expenses. This implies ha, possibly, hese funds are no passive porfolios and may serve purposes beyond hose of simply ransiional accouns. Deailed analysis of he daa is provided in Secion Facors influencing MMMF reurns Domian and Reichensein (1997) examine he facors ha affec he cross-secion of ne reurns of MMMFs, and he persisence of relaive reurns across years, finding ha 13

15 expense raio is he mos imporan facor in explaining difference beween ne reurns, and ha he MMMFs relaive reurns show srong persisence. I examine addiional facors and find ha i is no only expenses ha deermine cross-secional variaion of MMMF reurns. The primary argumen for observed differences in MMMF yields is ha hese funds pursue differen risk levels, as invesors and/or fund family may use MMMFs for differen purposes and may have differen risk preferences. DeGennaro and Domian (1996) examine ime-series differences in MMMFs average mauriy and conclude ha managers selec heir arge level for ineres-rae risk. The quesion hen is how managers decide wha level of risk hey are willing o ake, and wha securiies hey will use in heir porfolios o provide he reurn corresponding o heir chosen risk. The choice of he risk and he reurn is obained by selecing securiies wih differen reurns and mauriies. Reurns are a composiion of a risk free rae, defaul premium and mauriy premium. 2 Therefore, changing qualiy and/or mauriy of securiies can aler reurns. Adjusmen of mauriy premium has been addressed in previous sudies; however, hese sudies excep DeGennaro and Domian (1996) did no examine i as a choice of asse composiion, bu raher a response o expeced rae changes. Knez e al. (1994) idenify he common facors ha describe money marke securiies reurns by using boh hree- and four-facor models, which include: (i) he level facor, which represens movemens in yields; (ii) seepness, which represens changes in seepness of yield curve, i.e. relaion o mauriy; (iii) he Treasury facor, which capures 2 Which is a srucure of a deb securiy reurn composiion, and since MMMFs use only deb securiies in heir porfolios, I consider his srucure. 14

16 credi risk in issues credi risk in Treasury issues and, for privae issuer, i includes bank risk and firm risk. An addiional facor is a privae issue facor. Alhough common reasons for holding money marke funds are liquidiy and ransacion services combined, MMMF invesors may have addiional reasons for uilizing hese accouns. For hose concerned abou he safey of heir invesmen, i is appropriae o place heir money in accouns ha include a porfolio of governmen securiies, acceping lower reurn on he invesmen. 3 For less risk-averse invesors, higher reurns in MMMFs wih higher risk securiies are more aracive. As invesors choose equiy muual funds in differen caegories based on heir risk-reurn preferences, and funds offer hese choices, similarly, differen MMMFs are offered o saisfy unique demands. This variey comes from diverse clienele, and is refleced in variey of porfolio asse composiions. I es MMMFs for he facors deermining heir reurns and predic ha MMMFs wih higher risk and higher weighed average mauriy have higher reurns. Based on Knez e al. s (1994) reurn facors of money marke securiies, his predicion esimaes he second and hird pars of risk composiion in reurn risk (defaul) and mauriy premiums. I employ he following model, which includes cross-secional fund and economy relaed facors: r rf = α 1Risk 2Mauriy 3Expenses 4 logtnai, 5Inf + ε (1) where, dependen variable r is he yearly gross reurn of he MMMF i in year calculaed by annualizing monhly reurns repored in he CRSP muual fund daabase, and r f is he risk free rae available in he economy a he beginning of ime. Risk is 3 For exremely risk-averse invesors, he choice will no include muual funds, as banks offer money marke accouns wih FDIC insurance of up o $100,

17 measured as a monhly reurn s sandard deviaion of he MMMF i in year. Mauriy is a weighed average mauriy of securiies holdings of he MMMF i in year measured in days. Expenses and logtna are an expense raio and he logarihm of oal ne asses of he MMMF i in year, respecively. Inflaion in he economy, Inf, is calculaed as he change of consumer price index from December of year -1 o year. Finally, ε is he error erm. I expec β 1 and β 2 o be posiive, as he reurn should increase wih risk and mauriy premiums. The remaining variables are conrols and included for he following reasons. Expense raio is a proxy for beer managemen, and, as beer services cos more, I expec he coefficien of his variable o be posiive. MMMFs wih higher expense raios should have beer performance in he form of higher reurns relaive o oher MMMFs. logtna is a proxy for he economy of scale, and is coefficien is expeced o be posiive, as larger MMMFs can be more flexible in a choice of securiies mauriy daes and would be expeced o choose longer mauriies, ranslaing ino higher reurns. Thus, logtna may be a proxy for mauriy as well. 4 Inflaion is a macroeconomic facor, and is coefficien is expeced o have a posiive effec as a measure of he price change risk premium. Due o he Fisher s effec, risk free rae may no fully reflec inflaion, and here may be divergence beween hese wo indicaors, hus boh are included in he model. 4 To es his effec, I also run he model wih exclusion of logtna. 16

18 2.2. Family facors The majoriy of MMMFs are offered by complexes, i.e. fund families ha manage oher ypes of funds as well. 5 As par of he complex, funds ha have differen objecives, e.g. growh, income, bond and ohers, have diverse risks, and may have unique needs in erms of cash when hey face redempions or purchases. Those risks and cash flows may inerac among individual funds wihin he complex. Fund families can provide addiional benefis o invesors in he form of poenial for economies of scale and scope and also in erms of asse managemen, as hey have larger pools of managerial sources, disribuion exernaliies and beer research qualiies. A fund family also has is own objecive funcion o maximize, which is relaed o fees generaed from funds in he family. Thus, a fund family may engage in differen sraegies, such as cash managemen, risk managemen, and diversificaion, leading o a variey of srucures of he MMMFs wihin he complex Cash managemen Marke ransacions necessary o bring he fund cash level o he arge are no free. I may be cheaper for fund managers o ransfer cash and asses wihin a family hrough MMMFs, elevaing he cash managemen funcion from he fund o he family level. I is possible o have no liquidiy flows ino or ou of he family, and, herefore, no ransacion coss. Anecdoal evidence of such sraegy use is an example from Vanguard Funds family. In July of 2004, Vanguard launched an MMMF available only o 5 There are cases in which a family offers only one caegory of muual funds, including only MMMFs, e.g. Cenennial Capial Corporaion offers only MMMFs. Also, here are cases in which families do no have MMMFs. 17

19 Vanguard funds and cerain russ and accouns managed by Vanguard. 6 Thus, he family creaed a special MMMF for cash managemen purposes a he family level. The cash managemen sraegy can work as follows. To illusrae, le us suppose an equiy fund currenly holds a arge level of porfolio allocaion, including he level of cash, and, a he same ime, i faces cash inflow as i sells shares o new invesors. Before he manager of his fund will use he cash o adjus he fund porfolio holdings o he arge, he will pu he cash in money marke securiies. An MMMF of he same family, when i faces redempion ouflow, needs o sell money marke securiies o obain cash. The funds can fulfill heir needs by going o financial markes ouside he family, or hey can ransfer asses wihin he family, wihou incurring ransacion coss. The laer one should be preferred. Therefore, in his ype of ransacion, he MMMF s role is o fulfill cash needs, no only of individual invesors, bu also ha of oher funds in he family. However, his paern of fund flows is feasible if cash flows of he MMMF and he oher funds of he family are highly negaively correlaed or if he MMMF holds very shor mauriy securiies ha provide cash on regular basis, wihou a need o sell hem, and wihou incurring addiional ransacion coss. 7 Thus, he higher he flow volailiy of oher funds in he family, he higher he level of cash needed by he funds o mee heir liquidiy requiremens; and herefore, if MMMFs play he role of family cash ceners, he shorer he weighed average mauriy of a MMMF. 8 6 Vanguard Marke Liquidiy Fund, Semiannual Repor However, holding securiies wih very shor mauriy may inroduce addiional ransacion coss associaed wih reinvesing he cash. 8 Chordia (1996) ess he hypohesis ha cash and cash equivalens held by muual funds increase wih uncerainy abou invesor cash flows, and ha cash flow volailiy and finds ha cash holdings increase wih volailiy. 18

20 Depending on he level of cash flow correlaion among funds in a family, MMMFs can also play a differen role. In he firs scenario, all flows happen wihin a family, and no cash leaves he complex; hus, a closed sysem. For example, an equiy fund had ransferred some amoun o a bond fund, he bond fund had ransferred o some oher equiy fund, and he laer ransferred o an MMMF. In his case, when all cash flows can be offse, o avoid any ransacion fees, he complex should no sell and buy securiies ouside he family and all cash ransfers should happen hrough an MMMF. In he second scenario, here is an inflow o an equiy fund, and a manager of ha fund places incoming cash ino money marke securiies for some period of ime. 9 A he same ime, an MMMF of he same fund family experiences ouflow. If he MMMF acs as a cash cener, hen, insead of wo ouside ransacions, here will be only one, or none, as money marke securiies can be moved o he equiy fund in exchange for cash ha will be used o fulfill wihholdings from he MMMF s shareholders. If he inflow and he ouflow are perfecly correlaed in ime and absolue value, hen no ransacion is required. However, if he flows are no perfecly offse, bu a leas some of he flows are correlaed, hen a par of hem can be ransferred wihin he family, and he res will incur ransacion coss sill lower han in he case ha all ransacions are done ouside he family. However, if cash flows due o liquidiy or porfolio rebalancing are one-sided, hen flows of he equiy fund (or he MMMF) canno be offse and all of hese flows will be ransaced ouside he family, and, herefore, will incur full cos. When possible, using 9 For simpliciy, I ignore oher funds in a complex; his does no change he logic and oucome of he sraegy. 19

21 MMMFs as a cash cener for a fund complex may reduce ransacion coss, and herefore, increase proceeds o he family. Thus, families ha have funds wih higher flow volailiy face elevaed need in erms of cash managemen. As a resul, if families use MMMFs as cash ceners, I predic ha a family wih funds of higher cash flow volailiy will have MMMF(s) wih more volaile cash flows and, in he case when he family acively manages hese flows, shorer mauriy. As correlaions beween family funds flows and flows o MMMFs can be differen across families, he flow volailiies and he asse composiions of he MMMFs are expeced o differ, as are he exens o which MMMFs can be used as cash ceners by he families. The following models are employed: CFVolMMMF β = α CFVolFam 4 1 Famurnover 5 2 CFVolFam*CFCor logtnammf 6 logtnafam 3 CFCor + ε + (2) Mauriy = α CFVolFam β β logtnafam FamMMFCor 6 2 CFCor 10 CFVolMMMF 3 RevMMF/RevFam Famurnover 7 + ε FamilyHI 4 logtnammf 8 + FamilyDiver + (3) where CFVolMMMF is he volailiy of MMMFs cash flows in he family and CFVolFam is he volailiy of cash flows of he oher funds in he family calculaed as 10, 11 a sandard deviaion of monhly cash flows in year. 10 Cash flows o he MMMFs and o he family are calculaed using Sirri and Tufano s (1998) mehodology. I use monhly TNA and reurns o consruc ne cash flows. The flows are calculaed as: Flow = TNA TNA 1 *(1 + R ), where TNA is MMMFs or he res of he family i s oal ne asses a ime, and R is he MMMFs or he res of he family s value weighed reurns over he prior monh. 11 Chrisoffersen (2001) measures flows o MMMFs as a percenage change in asses, hough she indicaes ha defining fund flows as Flow = Asses Asses 1 * (1 + Ne Re urns ) / Asses does no change he 1 resuls of her sudy. Even hough Chrisoffersen s mehodology may be jusifiable for MMMFs, I use Sirri and Tufano s (1998) mehodology because I need o have consisen measures of fund flows for all ypes of fund invesmen objecives. 20

22 I predic a posiive relaion beween cash flow volailiy of MMMFs and of he family. If here is such a relaion, hen families, as well as invesors in he families, use MMMFs as cash ceners. In addiion, if mauriy is a ool o manage cash and liquidiy, hen i should be negaively relaed o he volailiy of cash flows of he family and he MMMFs. As volailiy of he family s cash flows increases, managers would shoren he mauriy of MMMFs securiies o release more cash for liquidiy purposes and o avoid addiional ransacion coss. CFCor is he correlaion beween ne cash flows o MMMFs and o he res of he funds in he family i. CFCor and CFVolFam* CFCor are conrol variables ha allow monioring as o wheher i is a closed sysem of cash flows. The res of he variables are conrols and explicily defined in he models for risk managemen predicions below Loads and cash cener The previous discussion abou he role of MMMFs in a fund family is from a family and managers poin of view, i.e. how managers can opimize ransacions in he family. On he oher hand, MMMFs can play a role for an invesor wihin he family. I is quie possible ha exisence of an MMMF wihin a family may play no role for an invesor who has a posiion in oher accouns of he family, as he invesor can liquidae he posiion whenever he wans o and place cash in a money marke accoun anywhere else ouside he family, or vice versa. If funds oher han MMMFs have fron and/or back loads, hen an invesor will face addiional expenses as a resul of moving his money in or ou of he fund. However, hese fees are omied if invesors asses are moved wihin a 21

23 family. Massa (2003) shows ha muual fund families employ sraegies ha rely on he heerogeneiy of invesors in erms of invesmen horizon by offering he possibiliy o swich across differen funds belonging o he same family, a no cos. Thus, for liquidiy purposes, he MMMF in he family can be more aracive han ouside money marke accouns o an invesor. Chordia (1996) develops a model and finds empirical evidence a he individual fund level, showing ha redempion raes are higher in funds wihou load fees han in funds wih fees and, herefore, cash holdings decrease wih load fees. I have a similar argumen, ha here is a higher need for cash a he family level for a high load fund family, as ransfers will be wihin he family, no ouside. Massa (2003) argues ha invesors who are planning o reallocae heir asses more frequenly will end o inves in funds wih lower load fees and in funds ha belong o larger families. I argue ha families wih higher average loads will experience more use of MMMFs by invesors, who realize he opion of a free move. I predic ha a family wih higher load funds includes a relaively larger MMMF(s), and a family wih higher load funds has higher volailiy of MMMF(s) cash flows. If invesors move heir asses among funds wihin a family, insead of hrough money marke accouns ouside he family, hen volailiy of he family s MMMFs cash flows is higher han for hose of a family ha does no resrain invesors from leaving he family by charging loads. Predicions are esed wih he following models, which are similar o Chordia s (1996) model of cash managemen ess a he fund level. The ess are conduced a he family level, where oal ne asses of MMMFs play a similar role in he family as asses invesed in cash in a single fund. 22

24 LIQ α 1CFCori, 2 AveFronLoad 3 AveBackLoad 4 logtnafami, + ε = (4) CFVolMMMF = α CFCor β 1 4 logtnafam 2 AveFronLoad LogTNAMMMF + ε 5 AveBackLoad 3 + (5) where LIQ is a family liquidiy raio. 12,13 AveFronLoad and AveBackLoad are he value weighed average of fron and back loads of he family i s funds in year, respecively. I expec loads o be posiively relaed o he family s liquidiy raio, and o he volailiy of he MMMFs cash flows. However, if loads discourage invesors from leaving he individual funds in he firs place, hen he family will have less need for cash managemen of invesors flows hrough MMMFs, as loads will fulfill ha role a he fund level. Thus, loads would decrease he family liquidiy raio in a manner similar o ha repored by Chordia (1996) a he fund level. I remains an empirical quesion as o wheher invesors are sensiive o loads and ake advanage of free asse ransfer opions wihin he family. The oher variables are conrols: he sizes of MMMFs and families, as well as cash flow correlaion may impac invesors decisions o use he opion of a free ransfer, or on a family s decision abou money marke securiies allocaion Risk managemen There is a pool of exising lieraure ha examines risk-aking sraegies of muual funds and risk sraegies wihin fund families. The laer sudies are of more ineres o his paper, and he main quesion ha hey explore is as o wheher fund families 12 LIQ = TNA TNA MMMFi, fam i, 13 Chordia (1998) measures liquidiy raio of a single fund as he cash and cash equivalens held by he fund as a percenage of he oal asses. My measure of family liquidiy raio is concepually he same. 23

25 maximize heir own objecive funcions raher han pursuing he bes risk adjused reurn sraegy for invesors. The agency problem creaes risk-aking behavior ha is no necessarily in he bes ineress of an invesor. Chevalier and Ellison (1997), among ohers, show ha a nonlinear convex shape of flow-performance relaion (Sirri and Tufano (1998)) creaes incenive for a fund manager o increase or decrease fund risk, which depends on he fund s year-o-dae reurn. Managerial fees revenues for a fund company depend on he oal asses under managemen, and, herefore, in order o maximize he fund objecive funcion, he fund manager has an incenive o ake acions ha increase fund inflows from invesors by changing risk, or using oher acions ha migh conflic wih invesors ineress. A conflic of ineress may exis as family affiliaions may influence he incenives of fund managers away from is shareholders ineress, if he whole family is going o benefi from a paricular sraegy. Gaspar, Massa and Maos (2006), and Guedj and Papasaikoudi (2003) argue ha families suppor beer performing and/or higher fee funds in order o maximize proceeds o he family and, herefore, o maximize he family s objecive funcion. This family sraegy of favoriism can be in he form of cross-fund subsidizaion, by shifing performance across funds (Gaspar e al. (2006)) as well as hrough limied resources allocaion across funds (Guedj and Papasaikoudi (2003)). However, family objecive funcion maximizaion can also be in a form of cash and risk managemen and diversificaion a he family level, which can reduce coss of running family funds, wih lile or no performance disurbance. 24

26 Massa (1998) develops a model of muual fund indusry srucure ha explains he role of fund families. He argues ha as consumers pick he funds on he basis of he whole bundle of services hey provide fund-managing companies behave as muliproduc firms. A fund family can hedge risk using caegory proliferaion as i makes he overall porfolio of he fund family more diversified. I can be argued ha, in an efficien marke, an invesor can achieve desired diversificaion on his own, so ha here is no need for a fund family o do so for him. However, wo explanaions for family diversificaion exis. The firs is ha families offer oher services ha make diversificaion for an invesor wihin a family more aracive han doing i on his/her own. The second is ha a family may wan o diversify in order o reduce is risk in a process of maximizing is own objecive funcion. Thus, o arac more invesors asses, families manage risk o capure a larger share of invesors wih various risk preferences, which is refleced in clienele effec as well, as discussed in he following secion. If a fund family comprises high-risk securiies, hen, o offse he risk, he family s MMMFs would be expeced o hold less risky and/or shorer mauriy securiies, and vise versa. I predic ha risk and mauriy of MMMF(s) of a fund family are negaively relaed o he risk of he family. Thus, differences in risk aking across MMMFs can come from diverse risk preferences of invesors, from cross-secional differences of an MMMF s reurn facors, and from risk diversificaion of a family porfolio. Tess are based on he following models conduced a he family level: Risk 1Familyrisk 2 Famurnoveri, 3 logtnammf 4 log = α TNAfam + ε (6) Mauriy 1Familyrisk 2 Famurnoveri, 3 logtnammf 4 log = α TNAfam + ε (7) 25

27 where, Risk and Mauriy are differen from he variables used o es composiion of MMMFs reurns. Risk is he sandard deviaion of monhly value weighed average reurns and Mauriy is he value weighed average mauriy of all MMMFs in he family i in year. Familyrisk is he weighed average volailiy of i family s reurns, measured by finding monhly weighed average reurns of he family porfolio, excluding MMMFs, and hen calculaing he sandard deviaion of hose reurns over he year. Famurnover is he value weighed average urnover of he family s porfolios, excluding MMMFs. logtnammf and logtanfam are he logarihm of oal ne asses of he MMMFs and of he family respecively, in year. Elon e al. (2003), Brown e al. (1996), Chevalier and Ellison (1999), and Kempf and Ruenzi (2004) use he volailiy of monhly reurns as a measure of risk. Chevalier and Ellison (1999) employ an ordinary leas squares mehod (OLS) wih risk as a dependen variable. I also use an OLS mehod on pooled daa o es risk managemen predicions. GLM procedure conrols for he family fixed effecs of he panel daa wih family dummy variables. If families use MMMFs as a ool o conrol family risk, hen for boh equaions, I expec β 1 o be negaive, as higher risk families will choose lower risk and lower mauriy MMMFs o diversify heir porfolios. Famurover is a conrol variable, and is higher value will require a lower mauriy of MMMF s securiies, in order o free some cash and reduce ransacion coss. logtnammf and logtanfam are conrols. Families have differen degrees of concenraion in a specific objecive ype, differen number of fund objecives, and a unique risk correlaion among funds in a family. For insance, a bond fund will have higher risk correlaion wih an MMMF, as 26

28 boh use deb securiies in heir porfolios and are dependen on yield srucure, and, hus, observe he same direcion of risks. In conras, if a family comprises mosly equiy funds, hen differen objecives wihin equiy funds may offse risks of each oher. Therefore, he role of an MMMF for family risk conrol purposes is more imporan in a single fund ype family, or in a more concenraed family, han for a family wih less risk correlaion. As a fund family has he desire o capure as much of invesors asses as possible, i would offer a diverse se of funds o cach invesors heerogeneiy. In a diversified family, invesors need no go ouside he family for diversificaion reasons. Khorana and Servaes (2004) find ha produc differeniaion is effecive in obaining marke share. Elon e al (2005) sugges ha a correlaion beween funds wihin a family may be higher han ouside he family, and ha he risk level in he family may be differen from wha can be obained from family diversificaion. To accoun for he level of in-family correlaion, diversificaion, and he risk correlaion of he oher funds wih MMMFs risk, I propose he following wih regard o he family risk managemen sraegy. A family wih higher concenraion in a single objecive caegory has an MMMF(s) wih lower risk and shorer mauriy. A family wih higher posiive correlaion in risk beween funds has an MMMF(s) wih lower risk and shorer mauriy. The following models are used o es he predicions. Risk = α Familyrisk β logtnafam 4 1 Famurnover FamilyHI 5 2 logtnammf FamilyDiver 6 3 FamMMFCor (8) β RevMMF / RevFam 8 Familyrisk * FamMMFCor 9 + ε Mauriy = α Familyrisk β logtnafam β Re vmmf / Re vfam FamilyHI 5 Famurnover 2 FamilyDiver logtnammf Familyrisk * FamMMFCor FamMMFCor 7 + ε + + (9) 27

29 where, Risk and Mauriy are a he family level, as defined in he previous es. FamilyHI represens family Herfindahl index, which measures concenraion of he family i in a specific objecive besides ha of is MMMFs. The Herfindahl index is defined as he sum of he squares of he family funds asses in each objecive caegory as a proporion of he family s oal asses. 2 N N FamilyHI i = TNA ji TNA (10) ji j= 1 j= 1 where, TNA ji is oal ne asses in fund s objecive j in family i and N is he number of objecive syles in family i. Based on he same reasoning for usage, and on closeness of he sample periods, I follow Massa (2003) and use ICDI_OBJ ou of hree poenial ses of caegories available in CRSP, which includes 23 differen objecives. FamilyHI i equals one if a family has only one objecive ype across is funds, and i is beween zero and one when a family has more han one objecive ype. Thus, he lower he value of FamilyHI i, he less concenraed he family. Anoher proxy for family diversificaion is FamilyDiver, which is defined here as one minus a sandard deviaion of he residual (σ ε ) from he Fama-French five-facor model, which capures idiosyncraic risk ha is no diversified away by a family porfolio. Thus, he larger he value of σ ε, he smaller he value of FamilyDiver variable, he less diversified he porfolio. The model o obain σ ε is as follows: 14 R RF = 1 RM RF ] 2SMB 3HML 4TERM 5 α [ DEF + ε (11) 14 Variables definiions are he same as in Fama and French (1993). The wo ou of seven Fama and French s (1993) bond porfolios used as dependen variables in he excess reurn regressions, are 5-year governmen bonds and corporae bonds raed Aaa. 28

30 FamilyMMFCor measures risk correlaion beween he family i s and MMMF s porfolios in year. I is a correlaion beween monhly reurns on MMMFs porfolio and he value weighed reurns on he family s porfolio, excluding MMMFs, in year. I also include as a conrol variable percenage of fee revenues generaed by MMMFs relaive o fee revenues of is whole family. These revenues are measured as an expense raio muliplied by TNA. Familyrisk*FamMMFCor is he ineracion erm included o conrol for he correlaion effec on family risk. The res of he variables are as defined before. I expec a negaive relaion beween family concenraion (FamilyHI ) and risk level and mauriy of MMMFs porfolios wihin he family. A posiive sign is expeced on he family diversificaion (FamilyDiver ) variable, i.e. more diversified families will require less risk managemen hrough MMMFs. FamilyMMFCor conrols for wheher here is room for risk managemen, and is expeced o be negaive, as lower correlaion will allow more possibiliies for using diversificaion sraegies Clienele For ransiional accouns and cash managemen, a family requires only one MMMF. However, in realiy, families have more han one (See Table 3). 15 The reason can be aribued o he clienele effec among invesors. Families have differen MMMF caegories and invesmen objecives, such as axable and ax-exemp, reail and insiuional. 16 Massa (2003) finds evidence of family-driven heerogeneiy among funds 15 For example, he Federaed Securiies Corporaion family has 105 MMMFs, including share classes. 16 Some examples of MMMF objecives, according o ICDI fund objecive codes include: MF Money Marke Tax Free Funds, which inves in municipal obligaions ha are close o mauriy; MQ High Qualiy Municipal Bond Funds, which inves in municipal securiies raed BBB or beer; among ohers. 29

31 and shows ha families acively exploi i. He argues ha when i is very cosly o compee on he performance dimension alone, he family will focus on oher ways of aracing invesors, such as by reducing fees or increasing he number of funds wihin he family. As invesors prefer differen risks and pursue differen sraegies in heir porfolios, hey may require differen levels of risk and reurn from MMMFs as well. Families ha pursue a sraegy of broad invesor coverage in erms of fund invesmen objecives should have more MMMFs wih differen invesmen syles. I predic ha he higher he number of fund syles wihin a family, he greaer he number of MMMFs wih differen characerisics and invesmen objecives ha are offered by ha family. This predicion is in line wih he family risk managemen sraegy, and is more pronounced for higher loads fund families. The higher he number of funds wihin he family, he greaer he value of he swiching opion, because he effecive fees decrease as a funcion of he number of funds. The OLS regression is esimaed based on he following models, conrolling for family fixed effecs: Dependen = α FamObjNum β AveBackLoa d 3 1 FamFundNum logtnammmf 4 2 AveFronLo ad logtnafam ε + (12) where proxies for dependen variable are: he number of MMMFs offered by he family i MMMFNum, he number of MMMFs invesmen objecives offered by he family i MMMFObjNum, and Herfindahl index of MMMFs in he family i MMMFHI. FamObjNum is he number of syle objecives in he family excluding money marke, and FamFundNum is he number of funds in he family i besides MMMFs. The res of he variables are as previously defined. For he dependen variables MMMFNum and MMMFObjNum, β 1 and β 2 are expeced o be posiive as increased number of fund 30

32 objecives and more funds wihin he family will indicae greaer invesor heerogeneiy, herefore, higher numbers of MMMFs wih differen invesmen syles will be required. For he dependen variable MMMFHI, he sign is expeced o be he opposie, as MMMFHI measures he concenraion, and is higher value indicaes less variey in he family s MMMFs. The signs on loads should be he same as in predicions for loads effec in cash ceners, indicaing invesors use of oher funds in he family. The remaining variables are conrols and are as previously defined. Two of he dependen variables are coun daa, and, o overcome resricion of OLS assumpion of coninuous normal disribuion of a dependen variable, I use he Poisson regression model wih hese variables and conrol for he family fixed effecs Family effecs In addiion o cash and risk managemen, and clienele predicions, which ry o explain cross-secional differences beween MMMFs, i is possible ha a family has oher specific effecs ha deermine differences. For example, he family can have generally higher expenses for all funds, and, herefore, MMMFs from ha family would have higher levels of expenses as well, compared o MMMFs from oher families. These higher expenses should be compensaed for by higher reurns, as funds ha show enhanced performance ha is achieved hrough beer managemen, require higher fees for ha experise. I predic ha he higher he average levels of expense raios for a family, he higher he expense raios of he MMMFs in ha family. An MMMF wih a higher 31

33 expense raio should show beer performance in he form of higher gross reurns relaive o oher MMMFs, conrolling for mauriy of underlying asses. A es of he firs par of his predicion is combined wih he reurn composiion es, as defined above. The second par is esed a he fund level, as follows: Expenses = α CFVolMMMF β LogTNAfam ε AveExpensesFam 2 LogTNAMMMF 3 + (13) where, Expenses is he MMMF s i and AveExpensesFam is he res of he family s value weighed average expense raios. Β 2 is expeced o be posiive, reflecing overall family sraegies in fee seings. The oher variables are conrols. 3. Daa descripion and empirical resuls The primary daa source for his sudy is he Cener for Research in Securiy Prices (CRSP) survivor-bias free US muual fund daabase. I limied he sudy period o because CRSP daa have many missing observaions prior o Ne asse value equal o one is used o idenify MMMFs. 17 I drop he fund observaions wih TNA less han $10 million, leaving 13,427 fund-year observaions of MMMFs. Descripive saisics of he money marke muual fund-year daa, yearly and over he enire period of he sudy, are presened in Table 1. The daa show ha here is subsanial cross-secional variaion among he MMMFs in he variables presened in he able. For example, he sandard deviaion of he weighed average mauriy is 17 days wih a mean (median) of 45 (46) days. A similar picure is observed for expenses: sandard deviaion is 0.29%, wih he mean (median) of 0.60 (0.59)%, and for he gross reurn: sandard deviaion is 1.72% wih he mean (median) of 3.75 (3.82)%. 17 I also check ICDI s fund objecive code and porfolio holdings o be fully invesed in cash. 32

34 Table 2 exhibis saisics based on he number of MMMFs across families, by year and for he enire sample period. Columns 2 and 3 repor oal numbers of families available in he muual fund indusry, and of he families ha have MMMFs, respecively. Alhough i may appear ha families wih MMMFs represen less han half of he number of all families, Table 3 shows ha, in erms of TNA, he families wih MMMFs are larger, and represen he majoriy of he muual fund indusry hey had more han 90 percen of he muual fund indusry asse share as of December I is noiceable as well ha he average number of MMMFs in a family has increased from four o almos en funds, and he median number has changed from wo o four funds per family, during he same ime period. The number of families ha offer MMMFs varies over he years, wih a peak occurring in 2000, which can be explained by waves in he economy and populariy of differen invesmen producs. 18 Mos of he ess used in his sudy are for fund-family relaions. Firs, using he lis of MMMFs ha I obained, I seleced all funds ha were in he same family as he MMMF. Some of he families ha were iniially seleced based on he presence of MMMFs did no have oher ypes of funds. Therefore, for he predicions ha require families wih funds oher han money marke, I drop he MMMF-only families from he sample. Oher sources of he daa are as follows. Fama-French hree facors are obained from Wharon Research Daa Services. Ineres raes of securiies wih differen mauriies and raings come from Federal Reserve Bank repors. Inflaion rae is calculaed from he consumer price index (CPI) as repored by US Deparmen of Labor. 18 The oal number of families was he larges in 2000 as well. 33

35 The firs se of resuls is from fund level ess, excluding family effecs. Some of he differences in he MMMFs characerisics can be explained by differing risk-reurn sraegies ha various MMMFs pursue. Table 5 repors he facors ha affec reurns for MMMFs. Coefficien on he Risk variable is posiive, as expeced, and saisically significan a less han 1-percen level, however, mauriy has almos zero effec. Coefficien on he LogTNA variable has a posiive sign, as expeced, confirming ha here is an economy of scale effec similar o bond funds, as repored by Philpo e al. (1998). Tesing he model wihou LogTNA does no confirm he assumpion ha he size of an MMMF can be a close proxy for mauriy, as he coefficien of Mauriy does no change much. Inflaion has a posiive sign as expeced. Expenses are posiively relaed o gross reurns, which are consisen wih he predicion ha funds wih higher fees should have higher reurns, as fees should reflec managerial abiliies. These resuls are saisically significan a less han 1-percen level. The cash cener predicion resuls are repored in Table 6. As expeced, cash flow volailiy of he family and MMMFs are posiively relaed a less han 5-percen level. Cash flows correlaion and he ineracion erm of he family cash flow volailiy and cash flow correlaion, used as conrol variables for indicaing an open or closed sysem, are posiively relaed o MMMFs cash flow volailiy, emphasizing he resul of he main variable. Thus, here is an indicaion ha families use MMMFs as cash ceners by clearing appropriae cash and securiies ransacions wihin a family hrough MMMFs. If money marke securiies mauriies were used as a means o conduc a cash managemen sraegy, hen a negaive relaion beween an MMMFs mauriy and family cash flow volailiy would be expeced. Resuls for CFVolFam variable in his model 34

36 are no significan. Thus, alhough families use MMMFs as cash ceners, he families do no use acive cash managemen sraegies by conrolling mauriy of MMMFs porfolios. Family cash managemen aciviies performed hrough MMMFs may have some effec on expenses in MMMFs, as addiional coss associaed wih hese aciviies may exis. I es wheher cash flow volailiy of MMMFs affecs heir expense raios. Resuls repored in Table 7 demonsrae ha expenses do no increase wih MMMFs cash flow volailiy. Thus, he benefis of hese sraegies may ouweigh addiional coss. 19 The level of MMMF asses across families is repored in Table 4. The percenage of he TNA of MMMFs in he TNA of he family, idenified here as a measure of he family s liquidiy, varies subsanially across families. Specifically, he mean is 34.86%, wih a sandard deviaion of 29.88%. Thus, here is cross-secional variaion in he level of cash allocaion a he family level, suggesing ha families have differen cash managemen sraegies depending on family characerisics. Resuls for load effec on family liquidiy raio and cash flow volailiy of MMMFs are repored in Table 8. Wih conrol for family fixed effecs, resuls indicae ha fron loads are posiively relaed o family liquidiy raio, as expeced. As invesors pay fron load fees only once a he enry o he family, hey choose o move heir asses wihin he family. Therefore, here is enhanced need for money marke securiies as ransiional accouns wihin he family, and MMMFs serve ha purpose for invesors. Back loads have differen effecs. Back loads are negaively relaed o family liquidiy raios. I is possible ha funds ha impose back loads arac invesors wih long-erm 19 However, even if here is a cos of running he cash managemen sraegy a he family level, i canno be passed on o individual invesors in MMMFs. As all family invesors enjoy he benefis, he coss can be refleced in he expense raios of all funds wihin he family. 35

37 invesmen horizons. Chordia (1996) suggess ha here is separaion beween invesors who rade in and ou of a fund ofen and hose who say for a long period of ime. Precisely, lower urnover invesors choose funds wih loads in order o avoid loss in value o acively rading invesors. Therefore, here is self-selecion among invesors. Thus, back loads arac long-erm invesors who do no inend o rade ou of he fund, and higher back loads discourage shor-erm invesors who migh use family cash ceners for rading purposes. The effec of loads on MMMFs cash flow volailiy is as follows. When loads are separaed ino fron and back, hen resuls are no conclusive, as hey lack saisical significance. However, oal family loads are posiively relaed o MMMF cash flow volailiy, which is consisen wih my predicions. As higher loads make i more aracive for invesors o move heir asses wihin he family hrough he cash cener, so he cash flow volailiy of MMMFs increases. Tess of he risk managemen predicions reveal he following picure. Resuls of univariae analysis of families, wih and wihou MMMFs, as presened in Table 3, indicae ha families ha do no have MMMFs have higher risk han hose families ha do. Specifically, he mean value of risk for he former is 4.2% wih a sandard deviaion of 2.9%, versus mean value of risk for he laer, which excludes MMMFs, is 3.1% wih a sandard deviaion of 2.2%. Even more, afer MMMFs are included ino he calculaion, mean value of a family risk becomes 2.2%, wih a sandard deviaion of 2.1%. Thus, here is self-selecion of he families in erms of risk, and families ha have MMMFs are less risky, using money marke funds o conrol heir risk. Differences in means are saisically significan, a less han 1-percen. 36

38 Resuls of regression analyses of he risk managemen sraegies are repored in Table 9. Panel A includes resuls for he firs proxy of a dependen variable sandard deviaion of MMMF reurns, which measures overall risk of he money marke porfolios. In Eq. (6), risk of MMMFs is posiively relaed o he family risk wih saisical significance a less han 1-percen level, which is opposie of wha was expeced, if families were o use MMMFs for risk managemen sraegies. One explanaion of his resul is ha families wih higher levels of risk choose higher risk invesmens for heir money marke funds as well, hough univariae analysis indicaes ha families wih MMMFs do have lower risk. As repored in panel B, wih mauriy as a dependen variable, he coefficien on family risk is negaive, as expeced, wih significance a 10-percen level for OLS regression. 20 Mauriy measures par of he overall risk and his resul indicaes ha families perform some risk managemen sraegies hrough adjusing mauriy of money marke porfolios. Resuls for he conrol variables have he following explanaion. Family urnover is negaively relaed o mauriy, indicaing ha as urnover increases, families conrol increased rading aciviy wih shorening mauriy of heir MMMFs. Size of MMMFs is posiively relaed o mauriy, indicaing ha larger money marke funds can afford o have longer mauriy for heir porfolios as hey may have more liquidiy, due o differing expiraions of heir holdings. The resuls of Eq. (8) and (9) ess for family concenraion and he level of family diversificaion are repored in boh panels under Model 2. I expec ha risk managemen sraegies are more ofen required for more concenraed and less diversified families, and are more feasible wih MMMFs whose risk is less correlaed wih he risk of he res of 20 Wih conrol for family fixed effecs, he power of some of he variables in he es diminishes. 37

39 he family s porfolio. Panel A repors resuls for he overall risk dependen variable, where he resuls for common variables are consisen wih hose of Eq. (6), as repored in Model 1. I find ha MMMF risk is posiively relaed o family diversificaion, as expeced, and is highly significan a less han 1-percen level; indicaing ha more diversified families have less need o employ diversificaion and risk managemen sraegies hrough MMMFs, and vice versa. Correlaion beween MMMFs and family risk is posiive, which is consisen wih he finding ha MMMF and family risks are posiively relaed. Panel B repors similar resuls for he oher risk proxy mauriy. The coefficiens of family Herfindahl index (FamilyHI ), diversificaion (FamilyDiver ) and risk correlaion (FamMMFCor ) have signs as expeced. 21 Concenraion of a family is negaively relaed o Mauriy, i.e. more concenraed families choose shorer MMMF mauriy o conrol for family risk. The correlaion beween MMMF and family risks is negaively relaed o MMMF mauriy, as expeced, indicaing ha for he lower correlaion families, use of MMMFs for family risk managemen sraegies is more feasible. Risk and cash managemen sraegies may affec he level of family reurns. I conduc univariae analysis of he value weighed family ne reurns for boh ypes of families wih and wihou MMMFs. This approach limis he abiliy o separae effec of hese sraegies on reurns, so I can conclude only abou join effec. Table 3 shows ha families wihou MMMFs, on average, have higher reurns han MMMF families, hough his may be due o he fac ha he former have higher risk in heir porfolios, and so are compensaed for ha risk. Indeed, he average level of risk for families wihou 21 However, saisical power is los for many variables when using he conrol for family fixed effecs. 38

40 MMMFs is 4.17%, versus he average level of risk of 3.05 (2.22)% for families wih MMMFs, excluding (including) MMMFs. Risk and cash managemen sraegies may generae higher risk adjused reurns. I can make some conclusions abou he effec of cash managemen sraegy by examining he levels of expense raios. Families ha have MMMFs, on average, have lower expense raios (1.00% including, and 1.14% excluding, MMMFs, versus 1.40% wihou MMMFs), which may be achieved by reduced ransacion coss. Resuls of he clienele ess are repored in Table 10. To check for he robusness of he resuls, I perform boh OLS and Poisson analyses wih family fixed effecs, he laer of which are specifically designed for coun daa ess. Resuls show ha here is a clienele effec in he families, which is refleced in he number of MMMFs and heir various invesmen objecives. The number of funds and he number of invesmen objecives in a family are boh posiively relaed o he number of MMMFs offered by ha family. Wih he use of he number of MMMF objecives as a dependen variable, he number of family objecives is also posiively relaed. Though, he oal number of funds offered by he family has mixed resuls, I can infer ha, as here is more invesor heerogeneiy in fund families, hose families offer more MMMFs of differen syles o mee a broad range of invesors needs. Use of MMMFs Herfindahl index as a dependen variable shows resuls consisen wih he above findings. Thus, he clienele effec is found o be presen. This paper s final se of analyses ess wheher family characerisics deermine cross secional differences beween MMMFs in erms of boh expenses and oher variables. Table 7 shows ha he family level of expense raios deermines hose of he 39

41 MMMFs wihin he family. The coefficien on he value weighed average expense raio of he family is posiive and highly significan, which is consisen wih my predicion. Also, as repored in Table 5, expense raios of MMMFs are posiively relaed o gross reurns of he funds, which is consisen wih he predicion ha higher fees resul from compensaion for beer managemen. 4. Conclusion In his paper, I examine cross-secional differences beween money marke muual funds during he period I find evidence ha fund families and heir invesors ofen use MMMFs as cash ceners, since family cash flow volailiy is posiively relaed o MMMF cash flow volailiy. In addiion, I discover ha loads affec a fund family s liquidiy raio, level of MMMF asses, and cash flow volailiy of MMMFs. Using an opion of free asse ransfer wihin a family, invesors assign MMMFs he funcion of a cash cener wihin he family, as well. The resuls of his sudy demonsrae ha boh fron and back loads play differen roles in discouraging invesors from moving asses in or ou, a he fund level, and, herefore, offer unique roles for cash managemen asks a he family level. Invesors in funds wih back-end loads end no o use MMMFs for ransiional purposes wihin he family. These resuls are consisen wih he self-selecion of invesors found in Chordia (1996), in which back-end loads are found o arac only invesors who do no inend o move heir asses, even wihin a family. In conras, fronend loads have he opposie effec, confirming my predicion ha fron loads have a posiive relaion wih a family s liquidiy raio. Toal loads are also posiively relaed o 40

42 an MMMF s cash flow volailiy. Thus, he cash managemen funcion is shifed o he family level. As a resul of analysis, I find ha MMMF reurns are deermined by risk facors, such as risk of MMMFs porfolios, expenses and size of funds, as well as macroeconomic facors. This provides a glimpse ino he roo causes of influences in he observed cross-secional differences beween MMMFs reurns a he fund level. In addiion, he resuls sugges ha fund families use MMMFs for risk managemen purposes. Univariae analysis shows ha families wih MMMFs have lower risk han hose wihou MMMFs. Using wo proxies for he risk measure sandard deviaion of MMMFs reurns and mauriy of MMMFs porfolios in regression analysis, I find ha MMMF risk decreases as families are less diversified and more concenraed. In addiion, I look for a clienele effec, in he effor o explain variey among MMMFs and heir invesmen syles across families. This variey is explained by he diversiy in he numbers and invesmen objecives of he family s oher funds. I can conclude ha families wih more invesor heerogeneiy offer more MMMFs, of differen ypes, o mee invesors needs. This is an indicaion of he presence of a clienele effec. Family characerisics also deermine cross-secional differences beween MMMFs wih regard o expenses. I find ha MMMF expenses are posiively relaed o he value weighed average expenses of he family, and ha expense raios of MMMFs are posiively relaed o gross reurns of he funds. Conrary o he percepion ha MMMFs are simply homogeneous ransiional cash accouns, his paper finds ha MMMFs play a larger role han one migh expec wihin a muual fund family. The characerisics of MMMFs differ subsanially in he 41

43 cross-secion, and hese differences can be explained by family-specific characerisics, including diversificaion and cash managemen sraegies a he family level. Applicaion of hese sraegies can reduce operaing coss and improve overall performance a he family level, which may ranslae ino significan invesor benefis. 42

44 References Brown, K., Harlow, W., Sarks, L., Of ournamen and empaions: an analysis of he managerial incenives in he muual fund indusry. Journal of Finance 51, 1, Carhar, M., On he persisence of muual fund performance. Journal of Finance, 52, Chevalier, J., Ellison, G., Risk aking by muual funds as a response o incenives. Journal of Poliical Economy, 105 (6), Chevalier, J., Ellison, G., Career concerns of muual fund managers. Quarerly Journal of Economics 114, Chordia, T., The srucure of muual funds charges. Journal of Financial Economics, 41, 3-39 Chrisoffesen, S., Why do money fund managers volunarily waive heir fees? Journal of Finance, 56, 3, DeGennaro, R., Domian, D., Marke efficiency and money marke fund porfolio managers: beliefs versus realiy. The Financial Review Vol 31 No. 2, May, Domian, D., Money marke muual fund mauriy and ineres raes. Journal of Money, Cred and Banking 24, November, Domian, D., Reichensein, W., Performance and persisence in money marke fund reurns. Financial Services Review, 6(3), Elon, E., Gruber, M., Blake, C., Incenive fees and muual funds. Journal of Finance 58, 2, Elon, E., Gruber, M., Green, T., The impac of muual family membership on invesor risk. Forhcoming in Journal of Financial and Quaniaive Analysis Fama, E., French, K., Common risk facors in he reurns on socks and bonds. Journal of Financial Economics, 33, 3-56 Ferr M., Oberhelman, H., A sudy of he managemen of money marke muual funds: Financial Managemen 10, Summer, Gaspar, J., Massa, M., Maos, P., Favoriism in muual fund families? Evidence on sraegic cross-fund subsidizaion. Journal of Finance 61, 1, February,

45 Guedj, I., Papasaikoud J., Can muual fund families affec he performance of heir funds? SSRN working paper. Invesmen Company Ac of 1940, Rule 2a-7 Kempf, A., Ruenz S., Family maers: he performance flow relaionship in he muual fund indusry. Unpublished working paper. Khorana, A., Servaes, H., Conflics of ineres and compeiion in he muual fund indusry. SSRN working paper Knez, P., Lierman, R., Scheinkman, J., Exploraion ino facors explaining money marke reurns. Journal of Finance 49, December, Massa, M., Why so many muual funds? Working paper INSEAD. Massa, M., How do family sraegies affec fund performance? When performancemaximizaion is no he only game in own. Journal of Financial Economics, 67, MFFB, Muual Fund Fac Book, Invesmen Company Insiue: Washingon, DC Packer, J., Pencek, T., Taxable money marke muual funds average mauriy index and shor-erm ineres raes. Papers and Proceedings, Academy of Economics and Finance, Fall, Philpo, J., Hearh, D., Rimbey, J., Schulman, C., Acive managemen, fund size, and bond muual fund reurns, Financial Review 33, Seyfried, W., Packer, J., Money marke muual funds and marke efficiency: implicaions for individual invesors. Business Ques 2001 Sirr E., Tufano, P., Cosly search and muual funds flows. Journal of Finance, 43, 5 (Ocober),

46 Table 1.1. Descripive Saisics of he MMMFs The able repors descripive saisics for 13,427 fund-year observaions of he MMMFs during he period. Daa source is CRSP muual funds. Variables: TNA oal ne asses measured in millions of dollars, Mauriy weighed average mauriy measured in days, Expenses expense raio measured in percenage, Reurn yearly gross reurn of MMMFs measured in percenage, and Risk sandard deviaion of monhly reurns. N is he number of funds. Year Saisics TNA, $ mln Mauriy, days Expenses, % Reurn, % Risk, % 1992 Mean N=679 Sd Dev Mean N=740 Sd Dev Mean N=665 Sd Dev Mean N=812 Sd Dev Mean N=991 Sd Dev Mean N=1,138 Sd Dev Mean N=1,206 Sd Dev Mean N=1,201 Sd Dev Mean N=1,219 Sd Dev Mean N=1,207 Sd Dev Mean N=1,168 Sd Dev Mean N=1,365 Sd Dev Mean N=1,036 Sd Dev Mean N=13,427 Sd Dev h Pcl h Pcl Median h Pcl h Pcl

47 Table 1.2. Number of MMMFs across Families This able presens he disribuion of money marke funds across families. Daa source is CRSP muual funds daabase. Variables: All families oal number of families in he indusry, Families w/mmmf number of families wih MMMFs. The res of he variables describe he number of money marke funds in a family. Year All Families Families w/mmmf Mean Sd Dev 25 h Pcl. Median 75 h Pcl

48 Table 1.3. Univariae Analysis of he Muual Fund Familys Risk, Reurn, Expenses, and Toal Ne Asses The able repors descripive saisics of muual fund families wih and wihou MMMFs during he ime period. Risk is measured as sandard deviaion of he monhly weighed average ne reurns of he family repored in he able as Reurn. Expenses are value weighed expense raios for he family. Risk, Reurn, and Expenses are expressed in percenage. Toal Ne Asses (TNA) is measured in millions of dollars. N repors he number of family-year observaions. N Mean Sd Dev Minimum 10 h Pcl. 25 h Pcl. Median 75 h Pcl. 90 h Pcl. Maximum Panel A: Muual Fund Families wihou MMMFs (µ1) Risk 2, Reurn 2, Expenses 2, TNA 2, Panel B: Muual Fund Families wih MMMFs, excluding MMMFs (µ2) Risk 2, Reurn 2, Expenses 2, TNA 2, Panel C: Muual Fund Families wih MMMFs, including MMMFs (µ3) Risk 3, Reurn 3, Expenses 3, TNA 3, Panel D: Difference in means Risk Reurn Expenses TNA µ1- µ *** 2.79 *** 0.39 *** -16,251 *** µ1- µ *** 1.30 ** 0.26 *** -12,039 *** 47

49 Table 1.4. Descripive Saisics of Families wih MMMFs The able repors descripive saisics of he families wih MMMFs during he ime period. Daa source is CRSP muual funds. Variables: TNAMMF/TNAFam is he raio, as a percenage, of he oal ne asses of a family s money marke funds o he oal asses of he family, i.e. a liquidiy measure of he family s money marke funds securiies holdings, RevenueMMF/RevenueFam is he percenage of fee revenues generaed by he MMMFs relaive o fee revenues of he oher funds in he family, FronLoad is he value weighed average fron loads of he family, BackLoad is he value weighed average back loads of he family, ExpensesMMMF is he value weighed average expense raios of he MMMFs in he family, and ExpensesFam is he value weighed average expense raios of he oher funds in he family, measured in percenage. CFvolMMMF and CFvolFam are volailiy of flows o MMMFs and o oher funds of he family respecively. CFcor is he correlaion beween cash flows o MMMFs and cash flows o he oher funds in he family. N repors he number of year-family observaions. N Mean Sd Dev Minimum 10 h Pcl 25 h Pcl Median 75 h Pcl 90 h Pcl Maximum TNAMMF/TNAFam 2, RevenueMMF/RevenueFam 2, FronLoad 2, BackLoad 2, ExpensesMMMF 2, ExpensesFam 2, CFvolMMMF 2, CFvolFam 2, CFcor 2,

50 Table 1.5. MMMFs Reurn Composiion The able repors resuls from esimaing pooled OLS and Fund and Year Fixed effecs regressions of he facors deermining reurns of MMMFs a he fund level for he sample of 13,427 fund-years over he period. The esimaed coefficiens are from regressions of he following equaion: ri, rf = α risk i Mauriy i Expenses i TNAi Inf 1, 2, 3, 4 log, 5 + ε where he dependen variable is MMMF s gross reurn minus risk free rae. The independen variables include risk of MMMFs measured as sandard deviaion of monhly reurns, mauriy of MMMFs porfolios measured in days, expense raios expressed in percenage, he log of oal ne asses of MMMFs, and inflaion. T-saisics are repored in parenheses. The symbols *, ** and *** indicae saisical significance a less han he 10-precen, 5-percen, and 1-percen levels, respecively. OLS Fund and Year Fixed Effecs OLS Fund and Year Fixed Effecs Inercep *** *** (-42.63) (-0.88) (-38.33) (-0.22) Risk *** *** *** *** (-7.02) (12.31) (-5.55) (12.33) Mauriy *** * *** * (5.50) (-1.72) (7.87) (-1.68) Expenses *** *** *** *** (9.74) (13.51) (4.13) (13.27) LogTNA *** *** (19.03) (3.27) Inflaion *** *** *** *** (33.25) (13.59) (32.50) (13.52) Number of obs. 13,427 13,427 13,427 13,427 R Adj R

51 Table 1.6. MMMFs as a Cash Cener of a Fund Family This able repors resuls from esimaing pooled OLS and Family Fixed effecs regressions of he family s cash flow volailiy on MMMFs cash flow volailiy and mauriy a he family level for he sample of 2,130 family-years over period. The esimaed coefficiens are from regressions of he following equaions: CFVolMMMF = α CFVolFam CFVolFam * CFCor CFCor Famurnover + Mauriy β CFVolMMMF 6 β logtnammf 5 = α CFVolFam 1 1 FamilyHI 7 logtnafam 6 2 CFCor 2 FamilyDiver 8 + ε Famurnover 3 FamMMFCor 9 3 logtnammf logtnafam Re vmmf / Re 5 vfam where dependen variables are cash flow volailiy and he value weighed average mauriy of MMMFs in he family. The independen variables include volailiy of family s flows (CFVolFam) and correlaion of flows beween MMMFs and oher funds of he family (CFCor), and heir ineracion erm. Oher variables are family average urnover, he log of MMMFs and he res of he family s TNA, family Herfindahl index (FamilyHI), family diversificaion variable (FamilyDiver) derived from Fama-French five-facor model, correlaion beween risk of MMMFs and of he res of he family (FamMMFCor), and he percenage of family s fee revenues from MMMFs (RevMMF/RevFam). T-saisics are repored in parenheses. The symbols *, ** and *** indicae saisical significance a less han he 10-, 5-, and 1-percen levels, respecively. CFVolMMMF Mauriy OLS Family Fixed OLS Family Fixed Inercep *** * (-12.78) (-1.83) (-1.08) (-0.08) CFVolFam *** ** (3.50) (2.38) (-1.23) (-1.22) CFVolFam*CFCor *** *** (3.44) (3.68) CFCor ** ** (2.00) (0.90) (1.29) (2.05) FamTurnover * (0.66) (0.68) (-1.87) (-1.22) LogTNAmmf *** *** *** *** (13.93) (4.59) (5.33) (4.38) LogTNAfam *** ** (1.37) (1.24) (-3.89) (-2.04) CFVolMMMF (1.24) (1.36) FamilyHI *** (-3.04) (0.46) FamilyDiver ** (2.38) (0.94) FamMMFCor ** *** (-2.17) (-2.54) RevMMF/RevFam (-0.79) Number of obs 2,130 2,130 2,100 2,100 R Adj R ε 50

52 Table 1.7. Deerminans of MMMFs Expenses This able repors resuls from esimaing pooled OLS and Fund and Year Fixed effecs of he facors deermining expenses of money marke funds a he fund level for 15,283 fund-years over he period. The esimaed coefficiens are from regression of he following equaion: Expenses = α CFVolMMMF β LogTNAfam ε AveExpensesFam 2 LogTNAMMMF Where CFVolMMMF is cash flow volailiy of he MMMF and AveExpensesFam is value weighed expense raio of he res of he MMMF family. T-saisics are repored in parenheses. The symbols *, ** and *** indicae saisical significance a less han he 10-precen, 5-percen, and 1-percen levels, respecively. 3 + OLS Fund and Year Fixed Effecs Inercep *** *** (38.40) (8.88) CFVolMMMF -4.76E-07 *** (-5.57) (-0.35) AveExpesesFam *** *** (36.90) (5.31) LogTNAMMMF *** *** (-40.79) (-7.32) LogTNAFam 7.29E (0.58) (-1.18) Number of obs 15,283 15,283 R Adj R

53 Table 1.8. Effec of Family Loads on he MMMFs Characerisics The able repors resuls from Family Fixed effecs OLS regressions of family average fund loads, excluding MMMFs, on he size of MMMFs in he family and cash flow volailiy a he family level for he sample of 2,151 family-years over he period. The esimaed coefficiens are from regressions of he following equaions: LIQ = α 1CFCori, 2 AveFronLoad 3 AveBackLoad 4 logtnafami, + ε CFVolMMMF = α CFCor AveFronLoad AveBackLoad + β logtnafam 4 1 LogTNAMMMF 5 2 where dependen variables are liquidiy raio of he family measured as oal ne asses of MMMFs relaive o oal ne asses of he family and cash flow volailiy of MMMFs in he family. Independen variables include correlaion of flows beween MMMFs and oher funds of he family (CFCor), value weighed average fron, back, and oal loads of he family and he log of oal ne asses of he family and of he MMMFs. T-saisics are repored in parenheses. The symbols *, ** and *** indicae saisical significance a less han he 10-precen, 5-percen, and 1-percen levels, respecively. + ε LIQ CFVolMMMF Model 1 Model 2 Model 1 Model 2 Inercep *** *** * * (5.75) (5.63) (-1.85) (-1.83) CFCor ** *** (1.06) (1.01) (2.53) (2.55) AveFronLoad ** (2.21) (1.24) AveBackLoad * (-1.88) (1.47) AveToalLoad * (1.19) (1.66) LogTNAFam *** (-4.20) (-4.93) (0.93) (1.07) LogTNAMMMF *** *** (4.74) (4.72) Number of obs 2,151 2,151 2,151 2,151 R

54 Table 1.9. Effec of he Family s Risk on MMMFs Risk and Mauriy The able repors resuls from esimaing pooled OLS and Family Fixed effecs regressions of he family risk on MMMFs risk and mauriy a he family level for he sample of 2,130 family-years over he period. The esimaed coefficiens are from regressions of he following equaions: risk = α familyrisk β FamilyHI 5 β familyrisk 9 1 Famurnover FamilyDiver 6 * FamMMFCor 2 FamMMFCor + ε 7 logtnammf 3 logtnafam Re vmmf 8 4 / Re + vfam where dependen variable is risk of MMMFs porfolio of he family. The independen variables include family risk measured as sandard deviaion of monhly value weighed reurns of he family s funds, family average urnover, he log of MMMFs, and he res of he family s oal ne asses (TNA), family Herfindahl index (FamilyHI), family diversificaion variable (FamilyDiver) derived from Fama-French five-facor model, correlaion beween risk of MMMFs and he res of he family (FamMMFCor), percenage of family s fee revenues from MMMFs (RevMMF/RevFam), and ineracion erm Family risk*fammmfcor. T-saisics are repored in parenheses. The symbols *, ** and *** indicae saisical significance a less han he 10-precen, 5-percen, and 1-percen levels, respecively. + Panel A: Dependen variable: MMMF risk Model 1 Model 2 OLS Family fixed OLS Family fixed Inercep *** *** *** (10.13) (1.46) (-3.67) (-3.13) Family risk *** *** *** *** (7.62) (8.72) (4.52) (5.47) FamTurnover (-0.46) (0.12) (-0.34) (0.38) LogTNAMMF *** (-0.36) (0.40) (-2.48) (-0.57) LogTNAFam ** ** (-0.42) (-2.61) (1.28) (-2.08) FamilyHI (-0.19) (-1.31) FamilyDiver *** *** (3.94) (3.60) FamMMFCor * ** (1.84) (1.95) RevMMF/RevFam ** (2.48) (1.18) Family risk* FamMMFCor (-0.28) (-0.41) Number of obs 2,130 2,130 2,100 2,100 R Adj R

55 Table 9 Coninued Mauriy = α familyrisk β FamilyHI 5 β familyrisk 9 1 FamilyDiver 6 * FamMMFCor Famurnover 2 logtnammf FamMMFCor + ε 7 3 logtnafam RevMMF / Re vfam where dependen variable is value weighed average mauriy of money marke funds in he family. The independen variables include family risk measured as sandard deviaion of monhly value weighed reurns of he family s funds, family average urnover, he log of MMMFs and he res of he family s oal ne asses (TNA), family Herfindahl index (FamilyHI), family diversificaion variable (FamilyDiver) derived from Fama-French five-facor model, correlaion beween risk of MMMFs and he res of he family (FamMMFCor), percenage of he family s fee revenues from MMMFs (RevMMF/RevFam), and ineracion erm Family risk*fammmfcor. T-saisics are repored in parenheses. The symbols *, ** and *** indicae saisical significance a less han he 10-precen, 5-percen, and 1-percen levels, respecively Panel B: Dependen variable: Mauriy Model 1 Model 2 OLS Family fixed OLS Family fixed Inercep *** ** (24.90) (2.48) (-0.35) (0.93) Family risk * ** (-1.72) (-0.74) (2.02) (0.76) FamTurnover ** * (-2.19) (-1.35) (-1.71) (-1.21) LogTNAMMF *** *** *** *** (9.37) (4.92) (5.90) (4.37) LogTNAFam *** *** (-3.17) (-1.05) (-4.50) (-1.57) FamilyHI *** (-3.41) (0.34) FamilyDiver (1.33) (-0.33) FamMMFCor ** *** (-2.13) (-2.64) RevMMF/RevFam ** (-2.19) (-1.18) Family risk* FamMMFCor ** (-2.29) (-1.46) Number of obs 2,128 2,128 2,100 2,100 R Adj R

56 Table Effec of he Family Fund Invesmen Objecives on he Number and Invesmen Objecives of MMMFs The able repors resuls from esimaing Family Fixed Effec OLS, and Family Fixed Effec Poisson regressions of family invesmen objecives on he number of MMMFs in he family and heir invesmen objecives for he sample of 2,152 family-years over he period. The esimaed coefficiens are from regressions of he following equaions: Dependen = α FamObjNum FamFundNum AveFronLo ad + β AveBackLoa d 3 1 log TNAMMMF 4 2 log TNAfam where dependen variables are number of MMMFs offered by he family MMMFNum, number of invesmen objecives of MMMFs offered by he family - MMMFObjNum, and Herfindahl index of MMMFs in he family MMMFHI. The independen variables include number of family invesmen objecives besides MMMFs (FamObjNum), number of funds in he family, excluding MMMFs, (FamFundNum), value weighed average of fron and back loads in he family, and he log of oal ne asses of MMMFs and he family. T-saisics for OLS and z-saisics for Poisson are repored in parenheses. Poisson regression repors Pseudo R 2. The symbols *, ** and *** indicae saisical significance a less han he 10-precen, 5-percen, and 1-percen levels, respecively ε MMMFNum MMMFObjNum MMMFHI OLS Poisson OLS Poisson OLS Inercep * *** (-1.91) (1.18) 3.95 FamObjNum *** * *** *** (-3.33) (1.86) (2.54) (0.77) FamFundNum *** *** *** ** (19.61) (8.22) (2.47) (0.11) 2.13 AveFronLoad * *** * (1.78) (0.51) (5.11) (1.14) 1.78 AveBackLoad * (-1.55) (-1.87) (-0.70) (-0.33) 1.30 LogTNA MMMF *** *** *** *** (10.83) (13.48) (19.32) (4.73) LogTNAFam *** (-1.30) (0.10) (-4.86) (-1.47) Number of obs 2,151 2,126 2,151 2, R

57 ESSAY TWO INNOVATIONS IN FINANCIAL PRODUCTS: CONVENTIONAL MUTUAL FUNDS VERSUS EXCHANGE TRADED FUNDS 1. Inroducion Innovaions in financial markes are imporan for marke developmen. A relaively recen example of innovaion comes from he muual fund indusry, wih he inroducion of exchange raded funds (ETF). These ETFs offer a claim on he same underlying asses as hose of convenional open-end muual index funds, bu have an organizaional form differen from ha of convenional muual funds and, accordingly, have differen feaures and oucomes for invesors. Some oher examples of compeing innovaions are money marke accouns offered by boh muual funds and banks, and fuures conracs wih he same underlying asses and/or rading on differen exchanges. All of hese innovaions were creaed o capure some par of he compeiive marke, and hough hey are cosly o develop, hey are beneficial o invesors, as hey add o he compleeness of he marke hrough increased liquidiy, ease of rade, possibiliies for hedging and arbirage, and addiional services. No all innovaions have been successful. For example, some sudies show ha he inroducion of redundan conracs in fuures markes have failed o arac enough marke share o survive (Duffie and Jackson (1989), Johnson and McConnell (1989), Silber (1981)). Successful innovaions also should be sudied o provide grounds for oher innovaions. No many empirical sudies exis on financial innovaions, as was poined ou by Frame and Whie (2002), and hose ha exis are clusered around very few producs. The main reason for his, as he auhors sugges, is unavailabiliy of daa. This 56

58 paper uses a broad sample of indexes o sudy he ETF as an innovaion in organizaional form and asse characerisics ha compee wih exising open-end index muual funds. Muual funds have exised for almos seven decades and offer a wide range of producs and benefis o heir invesors. Index funds represen abou $1.5 rillion of muual fund indusry asses. However, a decade ago, a new fund ype, he ETF, was inroduced. This produc became very popular, especially in he las several years, suggesing ha here was a room for improvemen on he exising index producs. If ETFs are more efficien, hen hey should gradually replace convenional index muual funds. However, his has no ye happened in realiy. Observaions show ha, even hough he producs seem o be he same: a reurn ha racks an index, he oucomes of invesing hrough a convenional index fund versus an ETF can be differen based on invesor-specific circumsances. This sudy addresses wheher convenional muual funds and ETFs are subsiues, and how heir coexisence in he marke can be explained. The main differences beween convenional index funds and ETFs are associaed wih he rading opions of fund shares for individual invesors, and fee and ax implicaions creaed wih hose opions. ETFs have lower fund level fees due o he eliminaion of individual invesor accouning by ETFs, as his funcion is shifed o brokers. Also, ETFs are perceived o be more ax efficien, as heir organizaional srucure allows hem o efficienly minimize capial gain disribuions. Daa analysis shows ha, on average, ETFs have smaller racking errors and are more efficien afer expenses. The resuls of he sudy demonsrae ha convenional and exchange raded index funds are subsiues, showing ha if ETFs are more efficien in erms of performance, 57

59 hen, aggregaed by index, one dollar of he flows of ETFs will ake abou 22 cens of regular funds flows. However, hough he fund level fees of ETFs are lower han hose of convenional funds, brokerage fees and commissions ha invesors have o pay for rading ETF shares may increase oal invesor expenses in cerain condiions. The coexisence of wo ypes of index funds can be explained by a clienele effec, which pus he wo ypes in differen marke niches. A es of clienele reveals ha ETFs and regular funds are no perfec subsiues, and each is preferred by a differen ype of invesor. Specifically, he resuls show evidence of a ax clienele, suggesing ha ETFs are chosen by ax sensiive invesors, while convenional funds may be preferred by ax exemp or insensiive invesors who value services of muual funds. A es for insiuional clienele did no reveal a significan difference beween he reacions of insiuional and reail invesors o he availabiliy of ETFs. The remainder of he paper is organized as follows. Secion 2 provides a lieraure review. Secion 3 compares index muual fund and exchange raded fund invesmens, examines he reasons behind invesors choices beween a convenional fund and an ETF, fund ype subsiuabiliy, and performance differences and characerisics. Daa and empirical analysis are given in Secions 4 and 5, respecively, and Secion 6 concludes. 2. Lieraure review An exensive pool of lieraure exiss on convenional muual funds, covering differen aspecs of he muual fund indusry, such as performance, managemen, and fund srucure (e.g., Sirri and Tufano (1998), Edelen (1999), Elon e al (1993), and Carhar (1997) among many ohers.). However, very few sudies are available on ETFs, 58

60 including hose ha compare convenional muual funds wih ETFs. The main reason for his is he shor ime period for which ETFs have exised, and he lack of availabiliy of relaed daa. Convenional muual funds were creaed in he 1940s and increased in populariy in he 1970s. The fis ETF was creaed in 1993, hough awareness and acive use by invesors have increased in recen years. The muual fund lieraure on index funds, specifically, he research ha looks a performance, coss, and racking errors, is relevan o his paper for analysis of boh convenional and exchange raded index funds. The main quesion raised in he lieraure is abou he radeoff beween enhanced reurns and he racking error of index funds. I is widely known ha socks added or removed from he S&P 500 index experience abnormal reurns on announcemen. Index fund managers could rade on he dae of announcemen insead of he acual change dae and enhance he reurn of he fund. However, ha would increase he racking error, and o achieve he low racking errors observed in pracice, an indexer mus closely follow he exac replicaion sraegy. Blume and Edelen (2003, 2004) show ha a sraegy for S&P 500 indexers of rading a he open following he announcemen of a change, raher han a he change, adds 19.2 basis poins o he reurn per year wih almos no added risk, bu wih subsanial increase in he racking error. Blume and Edelen (2003, 2004) argue ha his addiional reurn is a measure of he delegaion cos in monioring an indexer hrough a racking error, and show ha less han half of he sudied indexers always follow he exac replicaion sraegy, consisen wih he hypohesis ha hese indexers are rying o recoup some of hese delegaion coss. 59

61 Gasineau (2004) addresses a similar issue relaive o ETFs, examining he performance of index ETFs relaive o he benchmark and convenional index funds. Insead of focusing on he ax efficiency of he ETF srucure, Gasineau (2004) looks a he operaing efficiency of index funds. By employing he sraegy discussed in Blume and Edelen (2003, 2004), Gasineau (2004) shows ha convenional index funds ouperform by making ransacions a some ime oher han he marke close on he las day of formal index rebalancing. Gasineau argues ha a srucural weakness in curren ETFs is a par of he explanaion for heir underperformance, and argues ha a change in porfolio managemen policy will permi ETFs o perform in line wih convenional funds on a pre-ax basis and, presumably, ouperform hem on an afer-ax basis in he long run due o ax-efficiency of he ETF srucure. Elon e al. (2002) also invesigae he performance of ETFs relaive o he underlying index benchmark. Specifically, hey examine he characerisics and performance of SPDR (Spiders) ETF and find ha is ne asse value is kep close o marke price by is abiliy o creae and delee shares by inkind ransacions. The resuls of he paper show ha SPDR underperforms he S&P 500 Index by 28 basis poins and low-cos index funds by 18 poins. Elon e al. s (2002) explanaion is he los income caused by holding dividends received on he underlying shares in cash. However, in spie of he documened performance disadvanage, he ETF is sill considered o be an imporan insrumen because of is organizaional form. ETFs are widely discussed as a prooype for he muual funds of he fuure. ETFs seem o offer he benefis of boh open-end and closed-end funds. In paricular, hey rade close o ne asse value (NAV), and, like closed-end funds, hey offer he abiliy o ransac a he marke 60

62 price a any poin during he rading day. ETFs avoid he disadvanages of closed-end funds for which prices may deviae widely from NAV, and he disadvanages of open-end funds ha are priced only once a day, and, in addiion, ofen have resricions or minimum limis on sales and purchases by cusomers. Mos of he sudies, including hose menioned above, are based on a limied sample of funds, and he resuls are represenaive only of big funds ha have a significan share and presence in he marke. Tha is jusifiable and valuable, bu a broader sudy on more comprehensive daa is necessary o make more accurae conclusions. One of he goals of his paper is o conduc such a sudy. Anoher aspec ha has been examined in he lieraure is he impac of capial gains axes on afer-ax reurns for shareholders of convenional index muual funds. Ineresingly, he majoriy of hese sudies were conduced before ETFs were inroduced o he marke. Muual funds are organized in such a way ha hey pass hrough all capial gains o he invesors. The problem of capial gains realizaion wihin muual funds due o he rebalancing of exising porfolios and he ax implicaions on shareholders is widely recognized by boh academics and praciioners. One way o deal wih his problem is o implemen a special rading sraegy ha would offse realized capial gains. Dickson and Shoven (1994) examine he feasibiliy of managing open-end and closed-end S&P 500 index funds in order o defer ne capial gains realizaion. They show ha i is possible o incorporae cerain feaures of he U.S. ax laws, i.e., wash-sale rules and he offseing of shor-erm and long-erm capial gains and losses, for poenial improvemens in pos-ax reurns o he invesors engaged in ax minimizaion sraegies. However, acive managemen of index funds would deviae from funds objecives and impose he problem 61

63 of increased racking error, i.e. here is a radeoff beween reduced capial gains and increased racking error. The inroducion of ETFs o he marke alleviaed he problem of managing capial gains realizaion. Poerba and Shoven (2002) examine he percepion of ETFs being more ax efficien han radiional equiy muual funds by comparing he pre-ax and afer-ax reurns on he larges ETF, he SPDR rus, wih reurns on he larges equiy fund, he Vanguard Index 500 fund. The resuls sugges ha, beween 1994 and 2000, boh he before- and afer-ax reurns on he SPDR and Vanguard Index 500 fund were very similar. The argumen for ax efficiency is ha by reducing he ax burden on invesmens in corporae socks hrough ETFs relaive o invesmens in he same socks hrough equiy muual funds, ETFs may move closer o he consumpion-ax reamen of corporae capial income. Plancich (2003) looks a muual fund capial gains disribuions and he ax reform ac of 1997, which made long-erm capial gains less axable, and finds ha managers appear o il heir disribuions owards he long-erm afer The reason for hem o do so is o make heir reurns more aracive afer-ax and o arac more cash flows. This implies ha invesors are ax sensiive and prefer higher afer-ax reurns on heir invesmens. However, no exising sudy, o my knowledge, examines how he characerisics of exchange raded and convenional muual funds define heir roles in he marke, or addresses he prospecs of heir coexisence and fuure developmen. If hese wo kinds of index insrumens coexis in he long erm, despie heir similariies, heir inricae differences in characerisics and heir implicaions should explain his coexisence. For example, some of he reasons for using ETFs as opposed o convenional index funds can 62

64 be (i) ax efficiency and (ii) invesors need for immediacy and rading opions, which can be used for conrolling risk and shor-erm rading. A reason why boh ETFs and convenional index funds are found in he markeplace can be, for example, he resul of a clienele effec. The low rae of axable disribuions on ETFs may make hem more aracive o equiy invesmens ouside of ax-deferred accouns, such as IRAs or 401(k)s. A he same ime, some aribues of radiional equiy muual funds may make regular funds more aracive for reiremen accoun invesors. Also, ETFs may be a par of an emerging rend oward segmenaion of he muual fund markeplace, wih frequen raders segregaed ino producs differen from hose preferred by low-urnover invesors. The challenge of consrucing empirical ess of he clienele effec and marke segregaion is caused by he poor availabiliy of daa on individual invesor rades. Kosovesky (2003) compares wo mehods of passive invesmen using a model ha is helpful in examining major differences beween ETFs and index funds. This model is based on invesor rading preferences, ax implicaions, and oher characerisics. Kosovesky (2003) shows ha he key areas of differences beween he wo insrumens are managemen fees, axaion efficiency, and oher qualiaive differences. I ake hese differences ino accoun in my empirical ess. The inended conribuion of his paper is o conduc a horough comparaive sudy of he performance, cos, and efficiency of ETFs and index funds using a more comprehensive daa se, and o es for subsiuion and clienele effecs. 63

65 3. Comparison of convenional and exchange raded index funds 3.1 Hisory of ETFs The period of inroducion of ETFs o he marke corresponds o he period of research relaed o problems associaed wih non-radabiliy and he organizaional srucure of convenional muual funds. 22 Before ETFs of modern form where developed, some pioneer forms came o he marke. As described by Gasineau (2001, 2002), he hisory of ETFs sars wih Index Paricipaion Shares (IPS), which racked he S&P 500 index and were firs raded in The IPS was followed by Torono Sock Exchange Index Paricipaions (TIPs) and Supershares. The firs ETF o sar rading on Amex in 1993 was Sandard & Poor s Deposiory Receips (SPDRs), wih a srucure of a uni rus. Laer, oher exchange-raded index producs were developed wih a srucure similar o muual funds as opposed o uni russ. One of he earlies of his ype is World Equiy Benchmark Shares (WEBS) now ishares MSCI Series. Currenly, as of Sepember 2005, abou 180 ETFs are available o invesors (Invesmen Company Insiue (ICI) repor and efconnec.com). The funds are offered by en differen sponsor companies and provide a large variey of domesic and inernaional underlying indexes and asses. 3.2 Concepual differences beween ETFs and convenional open-end index funds The firs and he mos imporan difference beween ETFs and convenional openend index funds is ha ETFs are raded in he secondary marke a he price prevailing a ha momen, and no a NAV. ETFs can be purchased or sold a any ime during a 22 For example, research on capial gains realizaion and axaion of muual funds (Dickson and Shoven (1994), and Bluoin e al (2000)). 64

66 rading day unlike convenional muual funds, he shares of which can be exchanged direcly wih he funds only a he 4 pm NAV as deermined by he funds. This opion of inraday rading may no necessarily be valuable o every invesor; however, i may appeal o invesors who are concerned abou he abiliy o ge ou of a posiion before he marke is closed when prices are volaile. Primary marke ransacions in ETFs consis of in-kind creaions and redempions in large sizes. This is anoher imporan characerisic of he organizaional srucure of ETFs ha disinguishes hem from convenional funds. The abiliy o rade like socks in he secondary marke makes ETFs similar o closed-end muual funds, bu he feaure of in-kind creaions and redempions makes ETFs very disinc from all oher ypes of managed porfolios. This also allows ETF managers o deal wih he problem of premiums and discouns due o divergence beween price and NAV. The possibiliy of inraday creaions or redempions is a significan facor in mainaining ETF prices exremely close o NAV. Also, redempion-in-kind can improve he ax efficiency of ETFs, which is imporan for he majoriy of invesors. In conras, mos redempions of convenional funds are for cash, and, in he case of significan fund holder redempions, a fund is required o sell shares of he porfolio ha may have appreciaed from heir original cos. 23 This procedure can realize capial gains, which have o be disribued o all shareholders, and even coninuing invesors have o pay axes on hese disribuions. ETFs, however, ake advanage of a special ax reamen hrough redempion-in-kind, hus improving heir ax efficiency. In such a scenario, he low cos basis shares of each 23 Redempion-in-kind in convenional funds is allowed on large amouns wih a minimum of $250,000; however, funds are relucan o do so. In addiion, he majoriy of invesors have posiions smaller han he specified minimum. 65

67 sock in an ETF s porfolio are delivered agains redempion requess. Convenional funds, in conras, ry o sell heir highes cos basis socks firs, leaving he cos basis of he porfolios low and, herefore, making funds subjec o higher capial gains laer, e.g. in case a paricular sock leaves he index and he porfolio needs rebalancing. Wih ETF in-kind redempions, a fund porfolio has a relaively higher cos basis, which means ha acquired socks generae smaller gains when hey leave he index. There are wo ypes of capial gain ax liabiliies: when invesors sell fund shares (conrolled by he invesors), and for funds aciviies independen of invesor rading (no conrolled by he invesors). ETFs creae ax efficiency for he laer ype, making ETFs more aracive for ax sensiive invesors. Since ETFs are raded jus like any sock, ETFs and convenional funds also differ in disribuion channels, which is anoher imporan facor. The shares of an ETF mus be purchased hrough brokerage firms, which enail commission coss, such as brokerage fees and a bid-ask spread. In conras, convenional fund s shares can be purchased direcly from he fund. Therefore, shareholder accouning for ETFs akes place a brokerage firms raher han a he funds. Eliminaion of he individual shareholder ransfer agency funcion reduces operaing coss. 24 The expenses of ETFs end o reflec he cos savings on his funcion (see daa descripion in Table 3). However, even if operaing coss are reduced, an individual invesor may face differen marginal coss when invesing hrough ETFs due o brokerage fees, commissions, and bid-ask spreads. Thus, depending on he invesor s rading aciviy and he volume of rade, he coss and, 24 One of he radiional funcions of he muual fund ransfer agen is o keep records of fund posiion placemens, so ha ongoing paymens based on 12b-1 fees or oher markeing charges can be allocaed o he appropriae persons. 66

68 herefore, preferences of invesing hrough an ETF versus a convenional fund can differ among ypes of invesors. In addiion, ETFs can be purchased on margin and sold shor, and some ETFs have raded opions, which are no available for convenional muual funds. These feaures can be imporan for invesors who perform risk managemen, and may be especially useful for insiuional invesors who are looking o hedge large-sized conracs. In his case, ETFs are aracive because hey have a large variey in racked indexes, and, unlike fuures conracs, hey do no expire. On he oher hand, ETFs and convenional index funds have many similariies. Boh have operaing expenses, which reduce invesors reurns. Mos ETFs o dae have been designed o rack a specific marke index, similar o he way convenional index funds do. Boh ETFs and convenional index funds may experience racking errors in maching pre-ax reurns on heir racked indexes (Blume and Edelen (2003, 2004), Gasineau (2004), Elon e al (2002)). However, even hough ETFs and convenional index funds offer similar producs, he differences lised above sugges ha hey may be preferred by differen ypes of invesors. ETFs may be preferred by inraday invesors who demand shor-erm liquidiy or immediacy in rade, by long-erm invesors who buy in large amouns and seek lower managemen fees, by hedgers and speculaors because of opions raded on ETFs ha allow for minimizing exposure or maximizing profis hrough leverage, and by invesors who are ax sensiive due o he ax-efficiency of ETFs. On he oher hand, convenional index funds would be preferred by acive invesors who make many small purchases or 67

69 sales due o no commission coss, by hose who place less value on liquidiy or immediacy in rade, and by hose who are ax exemp or less ax sensiive. 3.3 Hypoheses formulaion Wha moivaes index invesors o choose eiher a convenional muual fund or an ETF? These funds provide he same produc in ha hey earn a reurn on some marke index, bu have differences in operaion ha can be advanageous for cerain ypes of invesors. Based on he predicions of economic heory, subsiues, complemens, and independen producs have differen quaniy reacions o price changes of oher producs. Fund fees, reurns, and racking errors are noable deerminans of invesors demand. Depending on he cross produc relaionship beween convenional index funds and ETFs, invesors would reac differenly o relaive variances in fee changes in convenional index funds and ETFs. If hese wo fund ypes are indeed good subsiues, hen due o he fee and rading advanages of ETFs, convenional index funds would be expeced o gradually disappear or lose a significan share of he marke. We do observe ETFs gaining marke share. However, he loss in he marke share of convenional index funds appears o be due o fund indusry growh, including growh in ETFs, bu no due o he ouflow of asses from convenional index funds (Fig. 1 and 2). Thus, even if hey are subsiues, his effec of a negaive flow relaionship beween he fund ypes may be diminished or emphasized by compeiive acions ha he funds may ake. Convenional index funds facing compeiion from ETFs are pressured o make adjusmens in operaions o mach he level of fees o ha of ETFs. For example, Fideliy 68

70 reduced fees in Ocober This was followed by oher funds (Economis, Sepember 2004). For ETF invesors, lower coss a he fund level may no necessarily ranslae ino lower coss a he invesor level due o brokerage fees, commissions, and bid-ask spreads, which may differ on an individual basis. 25 Non-price compeiion, on he oher hand, as refleced in differen organizaional srucures and services provided, may diminish he subsiuion effec by segregaing invesors ino differen niches. In his case, ETFs would no be expeced o compleely drive convenional index funds ou of he marke Subsiuion effec The analysis sars wih examining wheher convenional index funds and ETFs are subsiues, and wha implicaions his may have for developmen of he indusry. Based on he similariies of he underlying producs, hese wo ypes of index invesmen are expeced o be subsiues, which should be refleced in fund flow relaions. Hypohesis 1: If convenional muual open-end index funds and ETFs are subsiues, hen hey will have a negaive fund flow relaion. If he wo producs are subsiues, hen demand for hese producs and he level of he subsiuion relaion will be deermined by heir prices. In he fund indusry, demand can be measured by asses allocaed or fund flows, and he price can be measured by fund fees or reurns adjused for fees. Thus, in price compeiion, funds may eiher reduce fees or enhance heir performance hrough reurns. However, for index funds, an addiional measure of performance is a racking error. Therefore, wo main crieria for 25 Therefore, in addiion o comparing performance ne of fund fees, i is imporan o look a he invesor level accoun performance ne of all fees. 69

71 exchange raded and convenional index funds evaluaion are he size of he racking error and oal fees. The funds wih smaller value of boh of hese crieria will generally be able o arac more of invesors money. The organizaional srucure implies, and daa indicae, ha ETFs have lower fees. A deailed analysis of he racking error and fees is available in a laer secion of his paper. Besides he differences in prices, he wo fund ypes differ in organizaional srucure, rading, and ax implicaions. ETFs have some non-price advanages, which can enhance he subsiuion effec. However, if differen invesors value hese benefis differenly, i can also reduce he subsiuion effec, leading o a clienele effec, which may explain why convenional index funds and ETFs can coexis in he longer erm Clienele effec One of he advanages ha ETFs have wih heir organizaional srucure is ax efficiency. As ETFs can realize fewer capial gains, hey impose less ax on individual invesors. Oher hings being equal, ETFs should generally be preferred o convenional funds by ax sensiive invesors. Anoher advanage of ETFs is inraday rading a he prevailing price raher han a sale NAV and addiional rading opions like shor selling, margins, and, someimes, derivaives. Therefore, ETFs should generally be preferred o convenional funds by inraday acive raders, hedgers, and speculaors, because ETFs give more flexibiliy in rading and provide more opions for risk managemen. However, hese rading opions of ETFs do no come free: They involve brokerage fees and commissions, which may increase oal expenses for invesors. This presens a radeoff beween he added benefis of he srucure of ETFs and addiional coss relaed o hose 70

72 benefis. Various invesors may have differen break-even poins, and, herefore, he choices of fund ypes (regular or ETF) may be differen. Thus, coexisence of convenional and exchange raded index funds may be explained by he clienele effec. Hypohesis 2: ETFs are preferred by invesors wih higher liquidiy and rading needs and/or higher marginal axes. This segregaion of invesors is parially due o he exisence of brokerage fees on ETF ransacions for individual invesors. An imporan crierion ha invesors consider for heir asse allocaion is he cos associaed wih he invesmens. If an invesor has o make many small purchases, hen he oal cos of he invesmen may be high, even if ETFs fund level fees are generally lower han fees of similar convenional funds. For he long-erm invesor who plans o make one large lump-sum invesmen, an ETF may be a good choice, as ETFs annual fees are on average lower han hose of regular funds. This may be a clienele effec based on invesors ime horizon: Long-erm invesors prefer ETFs due o lower managemen fees, and shor-erm acive invesors prefer muual funds due o no commission coss. Anoher crierion ha invesors ake ino accoun is he ax consequence of he invesmen. ETFs are generally expeced o be more ax efficien han regular index funds due o heir organizaional srucure. Reiremen accouns are eiher ax exemp or ax deferred, and, herefore, invesors in hose accouns may no gain addiional value in ETFs ax efficiency. Also, insiuional invesors may no necessarily be very ax sensiive, because hey pass hrough heir ax liabiliy o individual invesors. 26 This is ax clienele, and invesors choices beween ETFs and convenional index funds are 26 Though, if an individual invesor, a final ax-payer, is ax sensiive, hen insiuional funds will end o be more ax sensiive o arac more flows. 71

73 expeced o be as follows. Tax sensiive invesors would generally prefer ETFs, while ax insensiive invesors would prefer convenional index funds for he addiional services provided. The nex crierion ha some invesors value in ETFs is he availabiliy of rading opions. Invesors more sensiive o volailiy, such as hedgers, speculaors, and inraday raders, would generally prefer ETFs. The majoriy of individual invesors do no represen his group of invesors. Insiuional invesors, however, can ake advanage of inraday pricing and rading opions, and are expeced o prefer ETFs. This leads o insiuional clienele. Supporing argumens for he above saemen follow. The dollar value of a ransacion for an insiuional invesor is usually high, which reduces brokerage fees as a proporion o he invesed amoun. In conras, reail invesors ransacions are smaller. In addiion o he common benefis ha ETFs offer o insiuional and reail invesors, insiuional invesors may beer benefi from he wide array of risk managemen and invesmen sraegies, such as equiizing cash, managing cash flows, equiy/fixed income asse allocaion, secor/counry exposure, hedge sraegies, relaive value and long/shor sraegies, and ransiions. As ETFs are expeced o be more suiable for insiuional invesors if ETFs and convenional funds are subsiues (Hypohesis 1), hen he subsiuion effec beween insiuional index funds and ETFs should be larger han he subsiuion effec beween reail index funds and ETFs. 72

74 4. Daa sources and descripive saisics Several daa sources are used for he analysis. The aggregae daa from ICI are used for analysis of indusry rends, and are presened in Table 1 and illusraed by Fig. 1. Overall, from 1993 o 2004, asses of equiy muual funds grew subsanially, indicaing an increase in invesors ineres in his ype of invesmen. During , equiy muual fund asses decreased; however, ETFs showed seady growh during he same period of ime. Beween 1993, he year he firs ETF was inroduced, and 2004, asses held by equiy muual funds increased almos six imes. Over he same period, ETFs grew from almos no asses o 5% of asses in equiy muual funds. Fig. 2 (obained from Economis prin ediion Sep 23 rd 2004) shows he marke shares of convenional index funds and ETFs for he ime period. The figure indicaes ha he share of convenional index funds remained almos unchanged from 1999 o 2004, saying a around 8%, and he share of ETFs increased significanly o abou 5% in The primary source of he fund level daa for boh convenional index funds and ETFs is he Cener for Research in Securiy Prices (CRSP) survivor-bias free US muual fund daabase. The complee daa se on ETFs is only available for he years afer Thus, I resric he sudy period o range beween 2000 and The sudy is done by maching ETFs wih convenional index funds racking he same indexes. The convenional index fund lis is obained from and he ETF lis is colleced from 27 From a universe of 180 ETFs and 369 convenional index funds, nine indexes racked by boh ypes of funds were idenified, giving a sample of 171 convenional index funds and 11 ETFs (see Appendix). 27 Wharon Research Daa Services refers o he same source used by researchers o idenify index funds. 73

75 The muual fund indusry aggregae daa come from he ICI websie. Index reurn daa are colleced from index providers. The sample is an uneven panel of monhly fund daa aggregaed by racked indexes beween 2000 and The number of funds per ime period may differ due o he inroducion of new funds. Table 3 presens he descripive saisics of ETFs and convenional index funds, separaed ino insiuional and reail funds, and grouped by index, during he ime period. The saisics indicae ha more reail han insiuional funds exis wihin each index group, and each group has more convenional funds han ETFs. Reail funds are, on average, larger han insiuional funds wihin an index group. ETFs are, on average, larger han convenional funds for six ou of nine indexes. Expense raios are he lowes for ETFs. Convenional insiuional funds are subsanially cheaper han reail funds. For mos of he indexes, ETFs did no have capial gains disribuions. Convenional index funds had capial gains disribuions averaging around $0.2 million per year. On average, flows o ETFs were posiive and subsanially higher han hose o convenional index funds. Reail funds experienced negaive average flows for several indexes during he sudy period. 5. Empirical analysis 5.1 Performance Before esing he main hypoheses, he performances of ETFs and convenional index funds are sudied. Due o variaions in organizaional srucure, a source of performance differences would come from differences in he abiliy o reac o index change announcemens and relaed racking error effecs (Blume and Edelen (2004), and 74

76 Gasineau (2004) among ohers). Performance differences also come from fund expenses, measured by expense raios. ETFs are expeced o have lower expense raios due o heir exempion from individual accoun bookkeeping. Recen developmens in he muual funds indusry show ha price compeiion exiss no only among convenional funds bu also beween convenional funds and ETFs. For example, in Ocober 2004, he Fideliy fund family reduced fees on is five main equiy index funds o a enh of a cen per dollar invesed (Economis, Sepember 2004). Oher funds follow similar sraegies o effecively compee agains comparable ETFs. These adjusmens in fees may be anoher indicaor ha convenional index funds and ETFs are, indeed, subsiues, and ha facing an increase in compeiion due o he inroducion of ETFs, convenional muual funds search for new ways o arac invesors. Performance ess are done by conducing univariae analyses of effeciveness and he racking error of regular index funds and ETFs. Effeciveness is measured as he difference beween fund reurn and racked index reurn. Tracking error is an absolue value of he effeciveness variable. Means are calculaed in each index group of funds, and he means of he wo groups are compared. Table 2 presens he resuls of he univariae analyses of effeciveness and racking error. Panel A shows he saisics for he differences beween gross fund reurns and index reurns and for absolue values of he differences. The means of hese variables are calculaed as he averages across he funds ha rack one of he nine sudied indexes. Then, -saisics are calculaed o es wheher he means are saisically differen from zero. Furher, a difference in means beween convenional index funds and ETFs is calculaed for each index and esed for saisical significance. For five ou of he nine indexes, convenional funds have posiive 75

77 means of effeciveness ha are significanly differen from zero, while ETFs show his resul for only one index. For wo indexes, Russell 2000 and S&P 500, he means of effeciveness are saisically differen for convenional funds and ETFs, indicaing ha convenional funds have higher effeciveness in gross reurns. The absolue value of effeciveness, i.e. racking error of gross reurns, is saisically differen from zero a less han he 1-percen level for fund ypes and for all indexes. The difference in racking error beween convenional funds and ETFs is posiive and saisically significan a less han he 1-percen level for all bu Dow Jones Indusrial indexes, indicaing ha ETFs generally rack underlying indexes wih gross reurns beer han convenional funds. Panel B presens similar saisics for ne reurns. On average, convenional funds underperformed four and ouperformed one ou of nine indexes, wih saisical significance a less han he 5-percen level. ETFs, on average, underperformed six and ouperformed one ou of nine indexes wih saisical significance a less han he 1- percen level. However, differences in he means beween groups of funds indicae ha he magniude of underperformance is smaller for ETFs for hree indexes a less han he 10-pecen level. The means for he oher indexes are no saisically differen from each oher. The racking error of ne reurns is saisically differen from zero for boh fund ypes and for all nine indexes a less han he 1-percen level. The differences in means indicae ha ETFs have smaller ne racking error for eigh ou of nine indexes a less han he 1-percen level. Convenional funds and ETFs have noiceable differences in sample size. To conrol for his, firs, he averages of ne reurns across funds are calculaed for each index, and index reurns are hen subraced, giving he effeciveness measure. Panel C 76

78 presens he resuls for his measure. On average, convenional funds underperformed hree and ouperformed one ou of nine indexes wih saisical significance a less han he 10-percen level. ETFs, on average, underperformed six and ouperformed one ou of nine indexes wih saisical significance a less han he 1-percen level. However, he mean difference of effeciveness beween regular funds and ETFs was saisically significan a less han he 5-percen level only for one index, and his was in favor of ETFs. Resuls for racking error are he same as repored in Panel B. Thus, i may be concluded ha on average, ETFs have smaller racking errors and are more effecive in reurns afer fees. 5.2 Subsiuion effec The bes way o es for he subsiuion effec beween index funds and ETFs is o look a he individual invesor level choices by following rades in personal invesmen accouns. As convenional index funds and ETFs buy similar underlying producs, we would expec ha boh ener an invesor s uiliy funcion. However, daa a he individual invesor level are no available. To overcome his problem, all invesors in he economy are looked a in aggregae and considered as a single represenaive invesor. Therefore, aggregae flows o convenional index funds and ETFs are used as he indicaor of he represenaive invesor choice, and he subsiuion effec beween funds is esed wih he following model of a sysem of equaions. FlowRF β 5 = α 1 lagindexre FlowETF 6 Re 2 lagflowrf lagre 7 lagflowetf 8 3 Expenses 9 4 LogTNA FlowIndusry + ε + (1) FlowETF = α 5 β lagindexre FlowRF 1 6 Re lagflowrf 2 lagre 7 lagflowetf 8 3 Expenses 9 FlowIndusry 4 LogTNA + ε + 77

79 where dependen variables are flows o regular funds (FlowRF) and o ETFs (FlowETF). Flows are measured using he mehodology of Sirri and Tufano (1998) and are calculaed as: Flow = TNA TNA 1+ R ), where TNA is fund i s oal ne asses a ime, 1 ( and R is he fund s reurn over he prior monh. Flow reflecs he dollar growh of a fund in excess of he growh ha would have occurred wih no fund flows and all dividends reinvesed. Explanaory and conrol variables include: lagged flows o ETFs and regular funds; lagged index reurn; curren and lagged reurn in regular funds and ETFs, calculaed as he value-weighed average across funds racking he same index; expense raio; and he log of TNA, also calculaed as he value-weighed average across funds racking he same index. Reurn, lagged reurn, expense raio, and he log of TNA on he righ hand side of he equaions are of hose funds whose flows are on he lef hand side of he model. Flow o indusry is measured as he sum of flows o equiy, bond, and hybrid muual funds ne of flows o index funds. If boh of he β 1 coefficiens are posiive, hen I canno rejec he hypohesis ha convenional index funds and ETFs are complemens. However, if eiher of he β 1 coefficiens is negaive, hen I can rejec he hypohesis and conclude ha convenional index funds and ETFs are subsiues. The res of he variables are for conrol purposes, wih he following expeced conribuions. Flows o he muual fund indusry indicae invesor senimen and level of indusry invesmen, and β 4 is expeced o be posiive. Lagged index reurn measures he araciveness of index producs. Curren and lagged reurns of a fund are performance measures used in equiy funds flow sudies (Sirri and Tufano (1998)). Expenses measure 78

80 invesors coss and are expeced o have a negaive relaion o he flows. Toal ne asses are used o conrol for a size effec. To es he subsiuion effec hypohesis, I use regular OLS. However, o conrol for he endogeneiy problem, where flows o regular funds and ETFs ener boh equaions as dependen and explanaory variables and may be endogenously deermined, I use he seemingly unrelaed regressions (SUR) approach. For boh OLS and SUR, I conrol for fixed effecs by including year and index dummy variables. Table 4 presens he resuls of he ess for subsiuion effec. Coefficiens β 1 on flows o regular funds and ETFs in boh equaions are negaive and saisically significan a less han he 1-percen level wih all es specificaions, i.e. OLS and SUR wih fixed year and index effecs. Therefore, I can rejec he null hypohesis ha convenional funds and ETFs are complemens and conclude ha hey are subsiues in aracing invesors flows. Also, resuls show ha fund flows are posiively relaed o lagged flows for boh convenional index funds and ETFs under he SUR fixed year effecs specificaion. This resul also holds for convenional index funds wih he oher model specificaions. Flows o convenional index funds are also posiively relaed o he indusry flows. Flows o ETFs are posiively relaed o fund reurns wih he significance level a less han 5 percen. SUR wih fixed year effecs indicaes ha flows o ETFs are negaively and flows o convenional index funds are posiively relaed o fund expenses, and boh fund ypes are posiively relaed o fund size. 79

81 5.3 Clienele effec Tax clienele To es he ax clienele hypohesis, I use a naural experimen of changes in ax law and in capial gains ax raes. Before 1997, any asse sold, regardless of he holding period, was axed as ordinary income subjec o a maximum rae of 28%. Afer May 6, 1997, he maximum rae on long-erm capial gains fell o 20%, while shor-erm capial gains disribuions remained axed a he ordinary income rae, which can be as high as 39.6%. The ax law change decreased he ax on long-erm capial gains and made hem more aracive, while he opposie happened for shor-erm capial gains. Mangers of muual funds have he abiliy o adjus rades in a way o realize long-erm gains insead of shor-erm capial gains. Plancich (2003) shows ha managers appear o have iled heir disribuions oward he long-erm afer Even if managers of index funds may no have as much flexibiliy when i comes o adjusing heir porfolios, hey sill can o some exen. Therefore, I can use his ax change even o es he hypohesis of he ax clienele effec. I expec ha, afer he ax changes of 1997, he subsiuion effec beween convenional index funds and ETFs should increase, as managers can make convenional index funds more aracive afer ax by managing capial gains disribuions. My daa sample is limied o , and his change in ax law falls ouside of he period of sudy. However, anoher favorable change in he ax raes on long-erm capial gains happened in May 2003: The maximum ax rae was reduced from 20% o 15%, while he ax rae on shor-erm capial gains remained unchanged. I use he following model o es for a ax clienele effec around his specific even wih a sysem of equaions. 80

82 FlowRF β 5 = α FlowIndusry 1 FlowETF 6 lagindexre 2 Afer*FlowETF 7 Re 8 lagflowrf lagre 3 9 Expenses 4 lagflowetf 10 LogTNA + + ε (2) FlowETF β 5 = α FlowIndusry 1 FlowRF 6 lagindexre 2 Afer*FlowRF 7 Re 8 lagflowrf 3 lagre 9 Expenses 4 lagflowetf 10 LogTNA + + ε where Afer = 1 for a period afer May 2003 and is equal o zero oherwise, and Afer*Flow is an ineracion erm ha capures he marginal effec of ax changes on flows o he funds. I expec β 2 in boh equaions o be negaive, indicaing ha he ax change and he resuling lower ax advanage of ETFs creae more of a subsiuion effec beween convenional index funds and ETFs. The β 1 coefficien is sill expeced o be negaive, showing a subsiuion effec beween he fund ypes. The res of he variables are defined as in earlier ess. Anoher way o es for differences in he ax clienele beween convenional index funds and ETFs due o ax efficiency is o include coninuous variables for ax raes or capial gains disribuions in he model. I expec he coefficiens on hese variables o be negaive, as increases in axes or capial gains disribuions make convenional index funds less aracive o non-ax exemp invesors relaive o ETFs. The following model wih a sysem of equaions is used o es his hypohesis. FlowRF = α β 5 1 FlowETF FlowIndusry 2 CapGainsRF 6 lagindexre CapGainsETF 3 7 Re 8 lagre 4 lagflowetf 9 LogTNA + + ε (3) FlowETF = α β 5 1 FlowRF FlowIndusry 2 CapGainRF 6 lagindexre CapGainsETF 3 7 Re 8 lagre 4 lagflowetf 9 LogTNA + + ε where CapGainsRF and CapGainsETF are value-weighed capial gains disribuions o regular funds and ETFs by index, respecively. The res of he variables are as defined in earlier ess. 81

83 Table 5 presens he resuls of he ess for ax clienele beween regular funds and ETFs. Panel A repors he resuls of he even sudy around he capial gains ax change. As expeced, coefficien β 2 is negaive and saisically significan a less han he 1- percen level in boh equaions. This indicaes ha as he ax advanages of ETFs over convenional index funds diminish, he wo become beer subsiues. The res of he variables show similar resuls, as in he previous model of he subsiuion effec. Panel B shows ha he capial gains disribuions of regular funds have negaive effec on heir flows, as was expeced Insiuional clienele The hypohesis regarding he insiuional clienele effec beween reail and insiuional invesors and he inensified subsiuion effec beween insiuional index funds and ETFs is esed using he same iniial model for he subsiuion effec, bu separaes he flows of regular funds ino subsamples of insiuional and reail funds. If ETFs are more suiable for insiuional invesors han for reail invesors, hen he β 1 coefficiens are expeced o be larger for he insiuional subsample han for he reail group. I perform an F-es o deermine wheher coefficiens in he wo subsample regressions are saisically differen from each oher. Since I use he SUR approach, here are some limiaions on he inclusion of all variables in one equaion in order o make a meaningful comparison of coefficiens across regressions. However, I run fixed effecs OLS on a model ha includes flows o ETFs as a dependen variable, and boh flows o insiuional and reail convenional funds as explanaory variables. The model is as follows: 82

84 FlowETF = α FlowIns β lagflowetf 5 β Expenses 10 1 FlowReail LogTNA FlowIndusry ε lagflowins 3 lagindexre 7 lagflowreail 4 Re 8 lagre (4) If here is a clienele effec beween insiuional and reail convenional funds, and if he former are beer subsiues for ETFs, hen, from his model, I expec boh coefficiens β 1 and β 2 o be negaive, bu β 1 o be larger in absolue value. Table 6 presens he resuls of he ess for insiuional clienele. Panel A repors he findings of he iniial model (1) for a subsiuion effec on wo separae subsamples. The coefficien of flows o ETFs wih he dependen variable of flows o insiuional funds is and he coefficien of flows o ETFs wih he dependen variable of flows o reail funds is , which are saisically significan a less han he 1-percen level. I could be suggesed ha insiuional funds may be beer subsiues for ETFs. However, he resul of he F-es shows ha only he coefficiens of lagged flows o regular funds are saisically differen from each oher across subsample regressions. Panel B presens he resuls from fixed effecs OLS, where boh flows o insiuional and reail funds are included in one regression as explanaory variables. I shows ha flows o boh ypes of convenional funds have negaive relaions wih flows o ETFs; however, he magniude of his relaionship is larger for reail funds, hough he economic difference is no large. 5.4 Summary of resuls The resuls of his sudy demonsrae ha convenional index funds and ETFs are subsiues. If ETFs are beer performers, hen one dollar of ETFs flows will ake abou 22 cens of flows from regular funds. If convenional funds are beer performers, hen 83

85 one dollar of regular funds flows will ake abou 1.3 dollar of ETFs flows. Compeiion beween he fund ypes mainly comes hrough fund expenses and he racking error. Univariae analysis shows ha ETFs expense raios and racking errors are generally lower han hose of convenional funds. Therefore, as beer performers, ETFs are gaining a share of he marke a he expense of regular funds, as Fig. 2 presens. However, if ETFs and convenional funds were perfec subsiues while ETFs suied invesor preferences beer, hen we would no observe hese wo ypes of funds coninuing o coexis. This paper shows ha ETFs and regular funds are no perfec subsiues, and clienele effecs exis beween he wo ha separae hem ino differen marke niches. By using an even sudy approach, I find evidence of ax clienele, suggesing ha ETFs are generally chosen over convenional index funds by ax sensiive invesors. A es for insiuional clienele did no reveal significan differences beween insiuional and reail invesors reacions o he availabiliy of ETFs. 6. Conclusion This paper examines one ype of financial produc innovaion and sudies how his innovaion influences invesors choices. The inroducion of exchange raded funds o he marke has been a successful innovaion, as refleced in he rapid growh of heir marke share and heir populariy in he invesmen indusry. I sudy how exising producs, convenional open-end index muual funds specifically, share he marke and compee wih ETFs. Similar o innovaions in oher invesmen producs, such as conracs in he fuures marke, he inroducion of ETFs has increased compeiion in he index fund marke. This has benefied invesors. I analyze wheher convenional index 84

86 funds and ETFs are subsiues and wheher differen feaures of hese funds creae clienele effecs, exending compeiion beyond prices. The sudy illusraes ha convenional index funds and ETFs are subsiues. However, inroducion of he new produc, he ETF, did no replace he exising produc, he convenional index fund. Raher, i creaed a new conrac ha added o he compleeness of he marke by offering new feaures previously unavailable in he regular funds. I find ha convenional funds and ETFs are close, bu are no perfec subsiues, as hey may be preferred by differen clieneles due o differences in he characerisics of he wo fund ypes. This innovaion is useful o boh invesors and he marke, as i creaes healhy compeiion in prices as well as service and produc feaures. 85

87 Appendix Index Convenional Funds ETFs Reail Insiue Barra Large Cap Growh Barra Large Cap Value Barra Small Cap Growh Dow Jones Indusrial Russell Sandard & Poors Midcap Sandard & Poors Smallcap Wilshire Sandard & Poors All indexes

88 References: Blouin, J., J. S. Raedy, and D. A. Shackelford. The Impac of Capial Gains Taxes on Sock Price Reacions o S&P 500 Inclusion. NBER working paper 8011(2000) Blume, M., and R. Edelen. S&P 500 Indexers, Delegaion Coss, and Liquidiy Mechanism. Rodney L. Whie Cener for Financial Research Working Paper No (2003). Blume, M., and R. Edelen. S&P 500 Indexers, Tracking Error, and Liquidiy: a Complex Answer o Profiing. Journal of Porfolio Managemen, vol 30 issue 3 (Spring) (2004), Dickson, J., and J. Shoven. A Sock Index Muual Fund wihou Ne Capial Gains Realizaions. NBER Working Paper 4717 (1994) Duffie, D., and M. Jackson. Opimal Innovaion of Fuures Conracs. The Review of Financial Sudies, Vol 2(3) (1989), Elon, E., M. Gruber, G. Comer, and K. Li. Spiders: Where are he Bugs? Journal of Business, vol. 75, no. 3 (2002), Frame, W., and L. Whie. Empirical Sudies of Financial Innovaion: Los of Talk, Lile Acion? Journal of Economic Lieraure, Volume 42, Number 1, March (2004), pp (29) Gasineau, G. Exchange-Traded Funds: an Inroducion. Journal of Porfolio Managemen, Vol. 27 Issue 3 (Spring) (2001), 88 Gasineau, G. Equiy Index Funds have Los Their Way. Journal of Porfolio Managemen, Vol 28 Issue 2 (Winer) (2002), 55 87

89 Gasineau, G. The Benchmark Index ETF Performance Problem. A Simple Soluion. Journal of Porfolio Managemen, (Winer) (2004) Invesmen Company Insiue. Muual Fund Fac Book. Washingon, D.C.: Invesmen Company Insiue (2001). Johnson E., and J. McConnell. Requiem for a Marke: an Analysis of he Rise and Fall of a Financial Fuures Conrac. The Review of Financial Sudies, Vol 2 (1) (1989), 1-23 Kosovesky, L. Index Muual Funds and Exchange-Traded Funds. Journal of Porfolio Managemen, Vol. 29 Issue 4 (Summer) (2003), Low Fideliy. Economis, Vol. 372 Number 8394, Sep , Poerba, J., and J. Shoven. Exchange Traded Funds: a New Invesmen Opion for Taxable Invesors. American Economic Review, V 92 iss 2 (May) (2002), Plancich, S. Muual Fund Capial Gain Disribuions and he Tax Reform Ac of Naional Tax Journal, Vol 56, No 1, par 2 (March) (2003), Silber, W. Innovaion, Compeiion, and New Conrac Design in Fuures Markes. The Journal of Fuures Markes, Vol 1 (2) (1981), Sirr E., and P. Tufano. Cosly Search and Muual Funds Flows. Journal of Finance, 43, issue 5 (Ocober) (1998),

90 Figure 1 Asses in Equiy Muual Funds and Exchange Traded Funds, Asses, billions of $ Equiy Muual Funds ETFs Domesic Equiy ETFs Years Source of daa: Invesmen Company Insiue Figure 2 Exchange Traded and Index Muual Funds Marke Share, end of year, % of oal Source: Sep 23rd 2004, New York. From The Economis prin ediion 89