How Smar is Exchange Inermediaed Money? Evidence from Lised Open-End Muual Funds in Europe William Li and Jerry T. Parwada School of Banking and Finance, Universiy of New Souh Wales Sydney, NSW 2052, Ausralia 15 December 2012 Absrac Tradiional muual fund disribuion channels are increasingly being complemened by he quoing of open-end funds on sock exchanges in Europe and, saring afer 2012, in Ausralia. This paper uilizes daa on European open-end muual funds ha rade on German exchanges o sudy differences in he behavior of exchange inermediaed flows and radiional flows. We empirically es wheher he well-known flow-performance relaionship holds for lised funds. We find evidence of a breakdown in he flow-performance relaionship for lised funds, especially for middle quinile performing funds. Flows ha originae from he exchange show performance predicabiliy conrary o money sourced hrough oher channels. The exchange rading channel is largely used as a disposal avenue for muual funds unis, causing a leakage of 7bn from he European muual fund indusry over he las decade. Overall, we find he exchange-raded channel brings ne benefis o invesors. Corresponding auhor. Email: j.parwada@unsw.edu.au 1
1. Inroducion Since he 18 h cenury, he creaion of muual funds has faciliaed he pooling of savings by invesors o access professional invesmen managemen and economies of scale. I was no unil 1924 1 ha he firs open-end muual fund allowed invesors o freely purchase and redeem heir invesmens, marking he sar of he open-end fund srucure. Tradiionally, open-end muual funds have been he dominan invesmen vehicles for reail invesors, pension and compensaion funds, due o is diversificaion benefis and wealh managemen funcions. Muual funds worldwide coninue o dominae he asse managemen indusry, as demonsraed by he fac ha he secor managed US$24 rillion worh of asses a he end of 2011, having peaked a US$26 rillion in 2007. Today, he indusry faces significan compeiion from cheaper and more liquid invesmen vehicles, mainly Exchange Traded Funds (ETFs), whose asses under managemen have grown by 55% since 2007 o surpass US$1 rillion in Ne Asse Value (NAV). While he ETF indusry is sill a small fracion of he muual fund indusry, invesors are growing more sophisicaed and are demanding lower fees, along wih more feaures from heir invesmens. To address hese issues, some European open-end muual funds are now radable on public exchanges in Germany. While his innovaion addresses compeiive issues such as fees and liquidiy, i raises quesions abou wheher he new disribuion channel encourages invesor behavior ha differs from ha exhibied when invesing is hrough he radiional over-hecouner channel. 2 This paper akes advanage of he adven of exchange rading for open-end muual funds o revisi he flow-performance relaionship. While muual fund flow behavior has been well researched since he early work of Gruber (1996), Sirri and Tufano (1993 and 1998), and Zheng (1999), laer inquiries have been concerned wih examining he issue in differen regions (see Oen 2002 and Ferreira 2012b) or under differen mehodologies (see Sapp and Tiwari 2006 and Frazzini e al. 2006). There is currenly very lile published empirical research 1 See Rouwenhors (2004). 2 Open-end muual funds raded on public exchanges are no o be confused wih exchange raded funds (ETFs), as we explain when we give insiuional deails in Secion 2.2. 2
ha separaes muual fund flows ino differen disribuion channels in order o sudy he flows separaely. A recen excepion is he concurren working paper of Sialm e al 2012 ha divides defined conribuion (DC) muual fund flows ino employer sponsored DC plan flows or radiional muual fund accoun flows. Our paper is he firs o examine he relaionship beween fund performance and he exchange channel of flows and o analyze he benefis i brings o invesors and he asse managemen indusry. The inroducion of exchange-raded open-end muual funds is perhaps he mos imporan variaion o he radiional disribuion channel since he incepion of daily rading in muual funds ha was imporan enough o warran sudies of he flow-performance implicaions such as Greene and Hodges (2001). Our aim is o es wheher he radiional convex performance-flow relaionship (Ippolio 1992, Goezmann and Peles 1997, and Sirri and Tufano 1993 and 1998) holds when flows occur hrough he new public exchange channel. This sylized arifac is imporan o revisi, as i is an indicaion of how resourceful muual fund invesors are, since he convexiy is believed o be he resul of invesors chasing pas winners bu failing o sell poorly performing funds. A possible explanaion for his irraional observaion can be due o behavioral biases such as he disposiion effec (Shefrin and Saman 1985). Hence, we challenge he null hypohesis ha all muual fund flows are idenical, regardless of disribuion channel. We organize our analysis o arge he exchange channel in wo ways. Firs, we compare he flow-performance relaionship for lised versus unlised funds. Second, for lised funds, we pariion flows ino hose from he exchange channel and compare heir flow-performance relaionship o ha of he radiional disribuion channel. The second secion of our analysis is possible because we are able o rack lised funds ino he Morningsar daabase and idenify heir aggregae flows and oher fund characerisics. Our resuls may be summarized as follows. We observe a convex flow-performance relaionship ha is similar o ha found in lieraure for unlised European open-end muual funds. However, we documen a very weak form of he radiional flow-performance relaionship, especially for he middle quiniles for lised funds. We also find ha flow- 3
performance oucomes change according o he performance measures - while boh lised and unlised funds show a srong linear relaionship beween flow and performance, he higher order relaionships for lised funds are inconsisen across differen performance measures. Regarding he pariioning of flows, we repor evidence ha he convex flow-performance relaionship holds for flows from he radiional channel, whereas he relaionship is almos non-exisen for exchange flows. Turning o performance predicabiliy, we show he smar money effec as described by Zheng (1999) especially applies o lised funds. However, similar o Sapp and Tiwari (2004), he smar money effec is absen afer conrolling for momenum. Unlised funds showed no performance predicabiliy a all. Afer pariioning flows beween exchange originaed and radiionally sourced money, we find evidence of performance predicabiliy for flows originaing from he exchange channel bu no for flows from oher channels. Taken ogeher, our resuls sugges ha exchange sourced flows emanae from more sophisicaed invesors. We rule ou he hypohesis ha he higher frequency rading enabled by exchanges exacerbaes he asymmeric performance-flow relaions. However, he daa a our disposal do no allow us o dig deeper ino he issues of invesor sophisicaion and knowledge. Ideally, o obain conclusive evidence on hese issues, we require access o muual fund clearing daa in order o idenify he direcion and counerpary for each rade. The resuls of our research offer several insighs. Firs, he evidence provides lessons o oher exchanges conemplaing he inroducion of a more ransparen and liquid plaform. For example, he Ausralian Securiies Exchange is in advanced sages of planning for he quoaion of muual funds on is equiy-rading plaform, saring in 2012. 3 Second, our sudy gauges he effec of lised muual funds on he European fund indusry and gives guidance o asse managemen companies o make decisions regarding he lising of heir funds on exchanges. Third, our findings have poenial implicaions for oher asses classes ha may be considered for 3 See hp://www.asx.com.au/resources/invesor-updae-newsleer/201203-asx-o-quoe- managedfund-prices.hm 4
public lising. In he pos global financial crisis world, regulaors are becoming increasingly ineresed in inroducing more ransparency and accounabiliy o differen asse classes. A proposed model is o allow more asses o be raded on public exchanges; hence his paper may provide preliminary guidance on how asses reac o public lising. Our paper conribues o exising sudies ha have focused on various muual fund characerisics such as expenses (Barber, Odean and Zhang 2005), fees (Chordia 1996, Huang e al. 2007a and Chevalier and Ellison 1997), performance (Grinbla and Timan 1992 and 1994, and Carhar 1997), and size (Chen e al. 2004, Khorana e al. 2005 and Yan 2008). We complemen he performance-flow lieraure ha differeniaes insiuional invesors from reail invesors and hose which find less convexiy among funds dominaed by sophisicaed invesors (Del Guercio and Tkac 2002 and Kaplan and Schoar 2005). The remainder of he paper is organized as follows: Secion wo presens he reader wih background informaion ino he sudy. Subsequen o ha, he paper is organized ino six differen secions and will be presened in he following order: review of exising lieraure, hypohesis developmen, daa, mehodology, resuls and conclusion. 2. Background 2.1 General Background In his secion we provide a primer ino he archiecure of open-end muual fund disribuion, focusing on he public exchange channel. The rading of open-end muual funds on exchanges is no o be confused wih he plaform for exchange raded funds (ETFs). Tradiionally, open-end muual funds are eiher disribued hrough financial inermediaries (e.g. financial planners or banks acing as brokers) or direcly wih he invesor. These radiional disribuion channels are slow and inefficien. In various markes, invesors are only permied o buy or sell heir muual fund unis a he close of business when fund managers calculae he fund s NAV. This misiming in calculaing he fund s price wih he closing price of he fund s asses has caused significan disadvanages for passive fund holders, as described by Greene and Hodges (2001) and Zizwwiz (2006). Moreover, Bergsresser, Chalmers and Tufano (2009) fail o 5
find any angible benefis from purchasing muual fund unis from brokers relaive o direcbough fund unis. Bergsresser, Chalmers and Tufano also find ha brokers have no more skill a asse allocaion han he average invesor. This evidence highlighs one of he bigges problems facing he muual fund indusry oday - he high cos of rading muual fund unis wih no addiional benefis for invesors. To reduce he cos of rading, new muual fund disribuion channels have been creaed, in paricular on proprieary and public exchange plaforms. In he las wo decades, wih a dramaic shif o faser elecronic rading for almos all financial asses globally, inermediaries have begun o offer inraday muual fund rading wih marke order books. This allows muual fund holders o buy and sell muual fund unis on an inraday basis, wih heir chosen financial inermediaries as heir counerpary, insead of waiing for he close of business window. In 1996, Berlin Borse inroduced open-end muual funds on is exchange using limi order books. However, as a resul of expensive fees and he absence of a muual fund clearing house, rading volumes remained low. In 2002, hrough he esblishmen of ClearSream s muual fund clearing services and he acquisiion of Vesima (a fully auomaed order roueing sysem ha enables fund orders o be processed wihou manual inervenion), Deusche Borse creaed a cenral European muual fund clearing house o provide invesors wih he opporuniy o rade muual fund unis on a cos effecive basis over he six German exchanges. The creaion of he exchange rading plaform and clearing house offers a complee sysem where invesors can inegrae heir muual fund rading aciviies ino heir day-o-day rading on exchanges. This also consiues a new muual fund flow channel, namely he exchange raded channel. The new exchange channel wih a limi order book arges more sophisicaed invesors ha rade via heir brokerage accouns as well as insiuional invesors, giving hem he abiliy o execue rades on an inraday basis raher han a he close of business. 6
In lae 2008, Deusche Borse appoined wo marke makers 4 on he Frankfur exchange o beer faciliae he flow of muual fund rading hrough he use of he coninuous aucion wih a limi order book. The XETRA plaform aims o mach he liquidiy in he open-end muual fund rading marke o ha of he Exchange Trading Funds (ETFs) marke. Wih en years of rading daa available, our paper is an exploraory analysis of his new muual fund rading mehod and how i compares o he radiional muual fund disribuion channel. In paricular, we are ineresed in wheher he exisence of he new exchange-rading channel is a benefi or a derimen o invesors and he muual fund indusry. 2.2 Insiuional Deails Before proceeding o formal analyses, we describe he insiuional seing and explain he difference beween open-end muual funds and ETFs. While boh ETFs and open-end muual funds are designed o pool savings ogeher for he purpose of invesmen, heir moneary and invesmen mechanisms are very differen. Firs, he invesmen goal is very differen. ETFs aim only o replicae he performance of heir designaed marke by replicaing he marke weighs, while open-end muual funds aim o bea heir designaed marke by aemping o allocae money ino fuure ouperforming asses in he marke. The second and mos imporan disincion is he moneary mechanism, which uilizes he cash in he funds o purchase asses on behalf of invesors. The selling and buying of muual fund unis direcly reflecs he AUM of he fund. When an invesor buys ino a fund, he fund uses his new inflow of money o purchase he underlying asse. Conversely, when he invesor sells ou of he muual fund, he fund would sell he underlying asse proporional o he invesor s holdings and use he proceeds o pay back he invesor. Hence, an invesor s acions have a direc impac on he demand and supply of he underlying asse in he marke. 4 Only Berlin, Frankfur, and Sugar have appoined marke makers for is muual fund rading plaforms. A deailed primer on open-end muual fund channel and rading can be found in he appendix B of his paper. 7
This focuses on he channels ha link invesors and he open-end muual fund marke. Figure 1 gives a visual represenaion of he differen channels ha are presen in he marke place. Open-end muual fund invesors have wo avenues o purchase fund unis. They have he choice of going hrough he radiional channel, which involves he advisor assessing he invesmen goals of he invesor hen presening he invesor wih a range of choices. If he invesor is saisfied wih he suggesions, he invesor s cash will be ransferred over in exchange for unis of he muual fund. Advisors usually have agreemens wih muual funds for allocaed quoas of unis (i.e. a pool of unis), and sales of hese unis are rewarded wih commissions. In a relaed sudy, Hackehal e al. (2012), using German brokerage accoun informaion, found ha wealhier, older and more experienced female invesors were more likely o use he radiional advisors channel. Alernaively, invesors have he choice of using an online broker, which has a direc feed from he specialiss a he exchanges. 5 The broker will hen display he opimal price for he invesor on he basis of he bes price execuion policy. Invesors also have he choice of going o a financial inermediary such as he local bank; however hese inermediaries are likely o pass on he clien s order o heir wealh managemen uni advisors or will direcly execue he clien s order on heir behalf, via he exchange raded channel wih he specialis. By conras, for ETFs here are marke makers who ac as inermediaries ha purchase he underlying asses from he marke. They hen place hese asses ino he ETF in exchange for ETF unis. These ETF unis are hen sold o he invesors using a bidding process on he exchange. When an invesor decides o sell he ETF unis, he/she simply sells he share back o he marke maker. The marke maker will hen price he uni and hen sell i off o oher invesors. The difference is ha as invesors dispose of heir open-end muual fund unis, he asses ha are represened by he unis are direcly sold o he marke; ETF unis remain securiized and are simply passed ono anoher invesor. The specialiss for open-end muual funds and he marke maker for ETFs perform similar roles. 5 The specialiss and he exchanges have he role of faciliaing he order book and ensuring marke liquidiy. An in deph primer on how he specialis conducs muual fund rading on exchanges can be found in he appendix of his paper. 8
3. Review of Exising Lieraure 3.1 Pariioning of Flows There is currenly very lile research published on which disinguishes beween he differen flow channels for muual funds 6. Exising sudies do no disinguish beween he differen channels of flow and only draw general conclusions abou muual fund invesors and he indusry as a whole. An excepion is he recen working paper by Sialm, Sarks and Zhang (2012), which dissecs flows ino differen sreams in order o examine he flow-performance behavior in deail; specifically flow behavior of company-sponsored defined conribuion (DC) plans versus nonsponsored DC ino U.S. muual funds. The auhors found significan differences in flowperformance sensiiviy and performance predicabiliy beween he wo ypes of flow. This indicaes ha here are varying idiosyncrasies wih differen ypes of muual fund flows. Sialm e al. s findings indicae ha our research opic is essenial in filling a void in curren research space. 3.2 Flow-Performance Relaionship The performance-flow relaionship in open-end muual funds is well documened. Early sudies by Ippolio (1992), Sirri and Tufano (1993) and Gruber (1996) observed a possible nonlinear relaionship beween flow and pas performance. Chevalier and Ellison (1997), applying semi-parameric mehods o es lineariy of he flow-performance relaionship, found weak evidence of non-lineariy. The auhors, suspecing a lack of qualiy daa for a robus conclusion, suggesed using a more sophisicaed piecewise linear model o es he funcional form of he flow-performance relaionship. Over ime, research has benefied from improved mehodologies. Sirri and Tufano (1998), Del Guercio and Tkac (2002) all documened he convex flow-performance relaionship where muual fund invesors are reurn chasing, bu fail o sell ou of poor performers. More recenly, a sudy by Ferreira e al (2012b) confirm his relaionship in many counries worldwide. 6 We noe ha here are sudies such as Keswani and Solin (2008), which looks a sources of muual fund flows bu no he channel in which he flows buy in or sell ou of muual funds. 9
The convex flow-performance relaionship is ofen explained in erms of behavioral biases such as he disposiion effec, i.e. he endency of invesors o sell winning invesmens early and hold on o losers for oo long. Shefrin and Saman (1985) proposed he anomaly and i was empirically confirmed by Odean (1998). US equiy invesors were found o no maximize heir uiliy according o convenional heory (see Kahneman and Tversky 1979 and Machina 1982), even afer conrolling for ax, rading coss, and he desire o rebalance porfolios. Shefrin and Saman aribued he occurrence of his anomaly o irraional invesor behavior such as seeking pride and avoiding regre, self-conrol and menal accouning. Research by Cici (2010), Bailey e al. (2011), and Singal and Xu (2011) found evidence of disposiion in US muual fund markes. Bailey e al. also found evidence ha reurn or rend chasing behaviors were relaed o behavioral biases, raher han he abiliy o observe managerial alen from pas performance. Raionaliy dicaes ha he flow-performance relaionship would no exis if invesors were capable of maximizing uiliy and exercise raionaliy. Raional invesors undersand ha pas reurns do no necessarily indicae fuure reurns. Reurn on is own should no be he single facor ha deermines managerial alen. However, as highlighed by he sudies menioned above, invesor behavior someimes becomes irraional and creaes rend-chasing behavior. If he exisence of he exchange-raded channel were o bring benefis o invesors, we would observe a reducion in reurn chasing behavior for lised funds as well as flows from he exchange-raded channel. 3.3 Performance Predicabiliy The abiliy of muual fund flows o predic fuure performance was firs documened by Gruber (1996), and Zheng (1999) and remains a holy debaed opic. Gruber and Zheng found ha fund flows in he shor erm exhibied performance predicabiliy, however Frazzini and Lamon (2009) discovered ha flow is dumb a longer horizons. Conrary o Gruber, Zheng and Lamon s findings, Sapp and Tiwari (2004) argue ha he smar money effec can be explained by momenum as documened by Jegadeesh and Timan (1993). Sapp and Tiwari argue ha muual fund invesors display no real fund picking abiliy, relaive o oher invesors, as hey 10
chase afer funds ha were recen winners, as described by he flow-performance relaionship. Using a UK daase, a recen sudy by Keswani and Solin (2008) found srong empirical evidence ha he smar money effec exiss for UK funds regardless of he source of flow (reail or insiuional). The auhors also commened ha he smar money effec is driven by fund purchases bu no wihdrawals. Separae o heir findings wih UK daa, by re-examining US muual fund daa on a monhly basis, Keswani and Solin (2008) found srong evidence of smar money effec in he US, even afer conrolling for momenum, in conrary o he findings of Sapp and Tiwari. 3.4 Persisence Evidence of muual fund performance persisence is abundan, Hendricks e al. (1993), Grinbla and Timan (1994), Brown and Goezmann (1995) and Carhar (1997) all found evidence of performance persisence in muual funds over he shor erm, whereas Grinbla and Timan (1992) and Elon e al (1996) s sudies demonsraed persisence in US muual funds. Conrary o recen findings, Jensen (1969) using an alpha measuremen mehod, found no evidence of persisence; alhough he samples used by Jensen and ohers differ in ime period. While he majoriy of persisence sudy is based on he US, Oen and Bams (2002) found very weak evidence of muual fund persisence in European muual funds, wih he excepion of hose locaed in he Unied Kingdom (UK). The abundance of muual fund persisence evidence indicaes ha major world muual fund markes exhibi signs of weak-form marke efficiency. While Carhar (1997) finds ha persisence in muual funds can be explained by momenum facors, funds ha adop momenum sraegies do no earn higher reurns. This invalidaes he idea ha pas reurns play a cenral role in muual fund managerial alen. The endency for performance o persis hen becomes a caalys in driving he flow-performance relaionship; i coerces invesors o assign a disproporional amoun of weigh o pas performance when evaluaing fuure fund performance and managerial alen. 11
3.5 Fees and Expenses Muual fund fees and expenses have received a wealh of aenion in boh research and in he indusry. Wih increasing compeiion from ETFs and managerial compensaion, muual fund fees and expenses have come under close scruiny. Recen research by Gil-Bazo and Verdu (2009) noed anoher puzzle in he muual fund indusry: he negaive relaionship beween fees and performance. Gil-Bazo and Verdu found robus evidence ha funds which charge higher fees generae disproporionally lower performance; he auhors aribue his anomaly o sraegic fee seings by muual funds in accordance o invesor performance sensiiviy. They also found evidence ha funds wih beer governance will iniiae more appropriae fee srucures. Furher re-enforcing his poin, Adams e al. (2012) found ha disproporionaely high fees are prevalen in funds wih poor governance srucures. If he exchange channels faciliaed a beer and more ransparen marke, we would see flows reac negaively wih expenses and he relaionship performance beween lised funds and fees as posiive. 4. Hypohesis Developmen As discussed in he previous secion, a very rich body of lieraure exiss on he convexiy of he flow performance relaionship. However, his body of lieraure does no disinguish beween he differen channels hrough which muual funds flows come; hence i reas all muual fund flows as a uniform variable. The availabiliy of publicly raded open-end muual funds allows us o dissec flow ino differen channels and o examine wheher well-esablished empirical findings sill sand for flows from differen channels. As a resul, we develop five hypoheses based on well-observed phenomena relaed o flows and performance. These hypoheses all aim o invesigae he benefis he exchange channel brings o he muual fund indusry from he perspecive of invesor behavior. Our firs wo hypoheses sem from he noion ha he exchange plaform allows invesors o purchase and dispose of muual funds unis faser han radiional channels. In oher 12
words, he exchange acs as an exension of radiional channels, which allows invesors o ake up posiions much faser han radiional mehods. Tradiionally, i is easier o purchase a muual fund uni han selling i, due o agency issues oulined by Chordia (1996) and Chevalier and Ellison (1997) whereby invesors are incenivized o purchase muual fund unis due o sraegic fee seing pracices by muual funds. Afer he invesor has purchased he muual fund uni (hence an inflow ino he fund), he manager will have every incenive o keep he invesor s money in he fund o increase annual fees colleced 7. Clauses, such as back-end loads, delays and complex procedures, coupled wih disposiion behavior of invesor creae disincenives o sell ou of underperforming muual funds. Wih he creaion of he exchange-raded channel, coupled wih a cenral clearinghouse, invesors are less resriced in he disposal of muual fund share unis. Convenience a a lower cos is a deliberae feaure of he design of he muual fund rading plaform, as refleced in he following quoe from a promoional brochure: The funds can be raded in real ime and subjec o no fron-end load on all rading days from 9 a.m. unil 8 p.m. Baader Bank AG. and ICF Kursmakler AG. have been commissioned as fund specialiss. They ensure he ongoing quoaion of he funds in real ime, hus guaraneeing liquid rading a he lowes ransacion coss, and he greaes possible ransparency for invesors. This allows privae invesors o benefi from he igh spread beween he bid and ask price. 8 Hence, we believe ha invesors are using hese newly developed plaforms o dispose of heir muual fund unis. This leads o our firs hypohesis. The confirmaion of he hypohesis provides suppor ha he muual fund rading exchanges bringing many benefis, especially in he form of less resricive rading mechanism for invesors, bu his comes a a cos for he indusry. Hypohesis I: Lised muual funds exhibi srong ne ouflows, relaive o unlised muual funds. 7 Fees ha are based on AUM no performance. 8 See Börse Frankfur funds available a: hp://deuscheboerse.com/mda/dispach/en/lisconen/gdb_navigaion/mda/200_marke_daa/100_s po_marke/conen_files/spo_marke_producs/mda_sp_boerse_frankfur_funds.hm 13
Wih he implemenaion of he exchange-raded channel, invesors now have he means o bypass radiional procedures and brokers for muual funds and hence hey should experience a reducion in heir rading coss. We heorize ha exchange flows would be less sensiive o expenses han oher flows due o a reducion in cos and he abiliy o sell muual fund unis wih fewer resricions. This creaes an environmen where he invesors can quickly sell ou of inefficien funds (i.e. high expense raio) as opposed o invesors who face complex disposal procedures of radiional channels. Hence we arrive a our second hypohesis. Hypohesis II: Exchange flows, relaive o oher flows, are less sensiive owards fund expenses, as measured by he expense raio Our hird hypohesis builds furher on he firs wo hypoheses. Wih improved access o selling, we should see invesors sell more underperforming funds i.e. here should a sronger ne ouflow of funds for underperforming funds. We also believe he exchange raded channel allows invesors o chase afer reurn more inensely han before, given he speed and frequency a which he invesor may rade, no o menion he fac ha invesors now have access o more frequen pricing and marke deph. Hence we come o our hird hypohesis. Hypohesis III: The opporuniy o conduc more frequen rades hrough he use of a public exchange will exacerbae he convexiy of he flow-performance relaionship. As a resul, we should observe a more convex flowperformance relaionship for lised funds in comparison o unlised funds. Leading on from our hird hypohesis, we believe ha flow-performance relaionship will remain non-linear as documened by Ippolio (1992), Sirri and Tufano (1993 and 1998), and Sialm e al (2012). Given ha Hypohesis hree predics a more convex relaionship for lised funds, we should observe a posiive quadraic funcional form which rends upwards. As for unlised funds, following Sialm e al. (2012), we believe he flow-performance relaionship o be in he form of a cubic funcion. Hypohesis IV: The funcional form of he flow - performance relaionship for lised funds- is in he form of a posiive quadraic funcion, bu for unlised funds, i is in he form of a posiive cubic funcion 14
Our las hypohesis examines anoher ineresing muual fund observaion: he smar money effec. As documened by Gruber (1996), Zheng (1999) and Frazzini and Lamon (2008), shor-erm muual fund flows are deemed smar if he flows are able predic fuure muual fund winners. Similar o our firs hypohesis, we believe he exisence of an exchange will exacerbae such an empirical observaion. While here are no sudies wrien implicily on hese muual fund exchange channels, we can look a sudies ha compare exchange rading and Over-The-Couner (OTC) rading of derivaives. Rosenberg and Traub (2006) and Swizer & Fan (2008) found ha exchange raded fuures conracs are preferred by informed raders, as hey conain unique informaion relevan o he price discovery process. Togeher wih he smar money effec, we believe ha he flow originaes from he exchange raded channel is likely o be hose from informed raders and hence we come o our fifh hypohesis. Hypohesis V: Flow from exchanges has greaer performance predicabiliy han flows from radiional channels. Our hypoheses are premise on he fac ha he exchange-rading channel forms an exension of exising channels. Hence, we should observe a more inense form of exising muual fund relaionships, especially beween flow and performance. 5. Daa 5.1 Muual Fund Characerisics Our firs daa source is from Morningsar Direc and is mainly concerned wih various descripive saisics of muual funds such as reurn, size, incepion dae, Toal Ne Asse values (TNA), fees, benchmarks, and caegory syle of he fund. These descripive saisics are sourced based on wo samples of European open-end funds, a lised and an unlised sample. These funds include funds ha are defunc, acive or aken over and hence, survivorship bias free. We acknowledge he exisence of he CRSP survivorship free muual fund daabase and is inclusion of defunc and merged funds o preven survivorship bias as oulined by Carhar e al. (2002). 15
However, as Elon e al. (2001) noed ha he CRSP muual fund daabase conains omission biases, which produces he same effecs as survivorship bias. Hence, we believe ha he Morningsar daabase is he mos appropriae daabase for his paper, given ha survivorship bias can be miigaed using a special reques o access funds ha have closed down. We also noe ha Elon e al. found ha he problems of survivorship and omission bias are more significan issues for older daa and small funds; bu given ha we are dealing wih large funds and more recen observaions, Elon e al. s observaion is less of a concern. Fama-French Facors and Benchmark s reurn daa are sourced from Kenneh R. French daa library and Thompson Reuers Daasream respecively. These benchmark and facor performance daa ses are hen merged wih raw reurn muual fund daa from Morningsar o creae benchmark and risk adjused reurns for performance calculaions. Descripive saisics are lised in Table 18 and 19. 5.2 Exchange Flow Muual fund rading daa is sourced from Securiies Indusry Research Cenre of Asia- Pacific (SIRCA), which in urn is direcly sourced from he Thompson Reuers Tick Hisory daabase. The daa covers every quoe and rade for muual funds from 1996 o June, 2012 across six Germany public exchanges; Berlin, Dusseldorf, Frankfur, Hamburg, Munich and Sugar. We exclude funds ha recorded less han five hundred rades per year from our sample. This is o ensure we capure funds wih sufficien liquidiy for he purpose of our empirical sudy. We use he Lee-Ready (1991) algorihm o classify he rade direcion of he daa from he six German exchanges. Table 3 summarizes he flow size and direcion passing hrough each exchange. However, Char 1 in he Appendix pains a more vivid picure of showing how he six German exchanges have been used as a selling channel for muual fund invesors, causing a leakage of ~ 7b from he European muual fund indusry over he pas 15 years. We also noe ha rading volume is paricularly low prior o 2002, due o he lack of cenral clearinghouses for muual fund unis. 5.3 Flow Definiion 16
We pariion oal flow (TFLOW ) ino wo differen sreams of flows, he exchange flow (EXFLOW ) sream and Oher Flow (OFLOW ) i.e. radiional flow sream. These differen sreams of flows reflec he channel in which muual fund flows occur. Firsly, we define TFLOW, which is he oal flow o fund i in monh which is calculaed as described in Chevalier and Ellison (1997) as well as Sirri and Tufano (1998): TFLOW TNA = TNA TNA 1 (1 + 1 R ) where TFLOW is fund i s oal flow a monh, and R is he fund s reurn over he previous monh. In oher words TFLOW represens he excess percenage growh of a fund over growh ha would of occurred naurally, as if here were no new inflows and ha all dividends have been reinvesed. EXFLOW is a measure of moneary flows in and ou of muual funds via he public exchange mechanism. Specifically, i is calculaed by he following: EXFLOW EXFLOW = TNA = TNA = d d n= 1 n, d n= 1 n, ( P ( P ( P n= 1 n, TNA TNA TNA TVOL 1 1 TVOL 1 n, n, (1 + R TVOL n, (1 + R ) ) ) TFLOW ) TNA ) TNA TNA 1 1 (1 + R ) where P n, is he price of he nh raded share of muual fund i during he monh, TVOL n, is he nh raded volume of he corresponding fund for monh. d represens he las rade for fund i during monh. OFLOW represens all flows excluding hose ha sems from he public exchanges, i.e. radiional flow. I is calculaed by he following: 17
OFLOW = d [ TNAi, TNAi, 1 (1 + Ri, )] ( P n= 1 n, [ TNA TNA (1 + R )] d ( P + = 1,. ),, 1 (1 n i TVOLi TNAi TNAi Ri = 1 TNAi, TNAi, 1 (1 + Ri, ) TNAi, 1 d ( P n= 1 TVOLi = 1 TNAi, 1. ) 1 TVOL n. ) TFLOW Boh exchange flow and oher flow are calculaed as a percenage of oal fund flow. This allows he magniude and direcion of he wo differen sreams o be compared on he same basis. All flow variables, syle, exchange and oher flow observaions are winsorized a he 2.5% level o accoun for he exreme values a he ail ends of he disribuion. Table 18 and 19 show ha he hree flow variables experience a high sandard deviaion and he validiy of some of he values are quesionable. While we were processing he daa, we observe obvious errors in raw daa poins downloaded from Morningsar. Hence, we believe ha a 2.5% winsorizaion o flow is appropriae o counerac some of he quesionable exreme values. Please noe ha we also performed all analysis based on daa ha was winsorized by 1%. The coefficien direcion and magniude were similar, wih only a sligh decrease in saisical significance. 6. Mehodology 6.1 Muual Fund Sample Selecion, ) (1) Two samples of muual funds are seleced on he basis of is presence on global public exchanges. Firsly, a lis of all funds raded on he six German exchanges is exraced and hen a sample of funds is obained afer removing hose ha are hedge funds, closed-end funds, fundsof-funds, propery funds, ETFs, and funds ha are domiciled in ax havens. This gives us a sample of 271 lised funds wihou any survivorship bias. The unlised sample of funds is filered op-down from he European open-end universe from Morningsar. Afer applying he same crieria for he lised funds and he addiional crieria ha he funds mus no be lised anywhere globally and have he righ o be markeed and raded across Europe or he world, obain a sample of 411 unlised European open-end funds, which are free from survivorship bias. 18
Table 1 shows descripive saisics on fund size and age as of June, 2012. The daa shows on average, lised and unlised funds end o be hose ha are larger and older. This does no come as a surprise, given ha he process of lising is rigorous and older and larger funds are more likely o arac larger flows given heir marke presence. We also noe ha Germany was under-represened in he unlised sample and ha Ausria, France, and UK was underrepresened in he lised sample. We aribue his o he relaive locaion of he fund s domicile and he geographical locaion of he exchanges. Given ha he six muual fund exchanges are locaed all in Germany, i is naural o expec more German funds lised on he exchanges (i.e. reducing heir presence in he unlised sample) due o proximiy and invesor demand; hence he above samples are in line wih expecaions. To saisically compare he wo samples, a -es is performed o deermine he significance of differences in he mean of each aspecs of he fund. Resuls can be seen in Table 2. 6.2 Trade Direcion Deerminaion Wih he lack of rade direcion from German exchanges; we use he Lee-Ready (1991) algorihm o classify he muual fund rades from he German exchanges. There are currenly hree ses of procedures ha have he abiliy o deermine rade direcion; he quoe rule, he ick rule and he Lee-Ready algorihm. The decision o apply he Lee-Ready algorihm is purely on he basis ha microsrucure research, such as Ellis e al. (2000), shows ha he Lee-Ready algorihm produces he mos accurae rade classificaion relaive o he quoe rule and he ick rule. However, Ellis e al. also documen he decrease in effeciveness of he Lee-Ready algorihm in classifying rades ouside he US. This reducion in effeciveness is also documened by Aiken and Frino s (1996) sudy on he ASX. Ellis e al. also provides evidence ha he Lee-Ready algorihm s accuracy declines in high frequency rading markes and markes wih complex archiecures. While hese claims are empirically esed o be correc, we believe ha he Lee- Ready algorihm is suiable for our sudy because of he similariy beween he marke in his sudy and he marke he algorihm was originally designed for. The Lee-Ready algorihm is designed for markes wih coninuous aucion presided over by a specialis, which is he same rading procedure and price deerminaion procedure in use oday on muual fund rading 19
plaforms in Germany. We also observe he lack of high frequency raders and large dominan block raders on he German exchanges for muual funds rades and hence, i is suiable o use he Lee-Ready algorihm o classify rade direcions. Table 3 summarizes he resuls of he Lee- Ready algorihm classificaion for our period of sudy. 6.3 Performance Measure We employ six differen performance measures in our empirical ess for robusness. These measures include risk-adjused facor alphas, caegory-adjused reurns, benchmarkadjused reurns, and raw reurns. The hree risk-adjused models used are Fama and French (1993) s hree facor model, Carhar (1997) s four facor model, and he Capial Asse Pricing Model (CAPM) one facor model: R R R Rf Rf Rf = β + β i = β + β i = β + β i m m m RM RM RM M, M, M, + ε + β + β SMB SMB SMB SMB + β + β HML HML HML HML + ε + β MOM MOM + ε (2) where R Rf are he monhly excess reurn of fund i over he risk-free rae and RM is he excess reurn of he marke porfolio over he risk free rae in monh i. The remaining facors are SMB i, HML MOM i ha respecively are he monhly size, value and momenum facors as described in Fama and French (1993) and Carhar (1997). The risk-adjus models are calculaed using a 36-monh rolling regression similar o a recen sudy by Ferreira e al. (2012a) and alpha is calculaed as he sum of he inercep of he model and he residual as in Carhar (1997). Alpha measures he excess reurn a fund manager is able o achieve adjusing for risk, size, value, and momenum; i.e. i measures he manager s abiliy o selec ouperforming socks and exploi marke imings. A posiive (negaive) alpha indicaes he manager is ouperforming (underperforming) he seleced benchmark. 20
The benchmark-adjused reurns are calculaed as raw reurn minus benchmark reurn, using he Morningsar assigned benchmark performance. A he absence of a Morningsar benchmark, we use he primary fund benchmark is used. Lasly, we calculae he caegory reurns by subracing Morningsar caegory reurns from raw reurns. Morningsar has niney-five differen fund caegories covering all asse classes. The creaion of he Morningsar caegories are a blend of boh he syle and asse class of he fund. 6.4 Flow-Performance Relaionship To examine he flow-performance relaionship, we deploy he following model, similar o ha of Sirri and Tufano (1998) and Ferreira e al. (2012b). While he exising lieraure uses a mixure of Fama-Macbeh and ime-fixed effecs models, we have been resriced from applying he Fama-Macbeh model due o low number of observaions a he beginning of our daa collecion. We also perform a Hausman es wih he null hypohesis saing ha he differences in he coefficiens are no sysemaic. The resul reurns a highly significan p-value of 0.0012, which srongly indicaes ha a ime fixed effec panel regression is he mos appropriae model for his daase. We noe he lack of he use of urnovers as an explanaory variable, his is due o daa availabiliy and suiabiliy of he variable. While urnover is used as a conrol variable in many muual fund sudies, we decided no o use urnover as a explanaory variable due o quesionable daa inegriy of he urnover variable from Morningsar. This is due o he difference in inerpreaion and calculaion across differen counries and asse classes as noed in Ferreira e al. (2012a). FLOW = β + 0 f ( Rank + β 30DVol 5 ) + β Size 1 1 1 + β SFlow 6 + β Age 2 1 1 + ε + β Exp 4 1 (3) We define Rank as he percenile performance rank of fund i relaive o oher funds in he same during monh, 30DVol -1 as he daily volailiy of fund i in he pas monh, SFlow -1 as 21
he flow ino fund i s caegory in he pas monh, Size -1 as he naural log of fund i s size a ime -1, Age -1 as he naural log of fund i s age a ime -1, and Exp -1 as he ne expense raio for fund i a ime -1. We execue he regression model as a year-fixed effecs regression wih clusering by fund sandard errors. The equaion above is he general model used o examine he oal flow-performance relaionship, we also use exchange flow and oher flows as he independen variable o closely examine he flow-performance relaionship of he wo channels. The following models acs as a replacemen of he independen variable, insead using exchange flow and oher flow, as deailed in secion 5.3. EXFLOW = β + 0 f ( Rank + β 30DVol 5 ) + β Size 1 1 1 + β SFlow 6 + β Age 1 2 + ε 1 + β Exp 4 1 OFLOW = β + 0 f ( Rank + β 30DVol 5 ) + β Size 1 1 1 + β SFlow 6 + β Age 2 1 + ε 1 + β Exp 4 1 I is well documened ha he flow-performance relaionship (Chavelier and Ellison 1997, Sirri and Tufano 1998) exhibis non-linear forms. Hence, we esimae he funcional form of he relaionship in wo differen ways. The firs funcional form divides flow performance ino hree linear secions base on performance. The second esimae implicily ess he funcional form of he flow-performance using linear, quadraic and cubic funcions. Following Sirri and Tufano (1998), we fi he percenile ranking of he funds ino hree differen linear performance secions; hey are calculaed using he following: f 1 ( Rank f, ) = γ L LOW f, + γ M MID f, + γ H HIGH f, (4) where LOW MID f, f, HIGH = MIN ( Rank = MIN ( Rank f, = Rank f, f, f,,0.2) LOW LOW f, f,,0.6) MID f, 22
The gammas capure he coninuous piecewise linear flow-performance sensiiviy, relaive o heir respecive performance quiniles. To calculae he rank variable, we rank each muual fund reurn in he sample on he basis of raw, facor adjused and benchmark adjused agains oher funds in he sample. We choose o focus our inerpreaion on he ranking of raw reurns relaive o all oher funds in fund i s Morningsar caegory. We define caegorical rank as he fund s raw performance rank relaive o oher funds in he same Morningsar caegory, and is being uilized as he main ranking variable. We believe ha caegorical rank is he mos robus and comprehensive way o rank he raw oal reurn of a muual fund given he size of our selecive sample. The second specificaion ess he flow-performance relaionship o see wheher i is a linear, quadraic or cubic funcion. f 2 3 2 ( Rank f, ) = γ1( Rank f, 0.5) + γ 2( Rank f, 0.5) + γ 3( Rank f, 0.5) (5) The gammas are he flow-performance sensiiviies for heir relaive specificaions. The equaion allows us o examine wheher he relaionship is linear, quadraic, cubic or conforms o specificaions of higher order and ulimaely draw conclusions on he concaviy of he relaionship. 6.5 Performance Predicabiliy The relaive sophisicaion of rading muual funds on exchanges leads us o believe ha he wo sreams of flows, - exchange and radiional may exhibi differen performance predicabiliy properies. Hence, we use he following model o examine predicabiliy of exchange flow and oher flow separaely. perf = β + + β OFLOW 0 2 + β AGE 5 1 1 + β perf + β EXP 6 3 1 1 + ε + β SIZE 4 1 (6) perf = β + β EXFLOW 0 1 + β AGE 5 1 1 + β perf + β EXP 6 3 1 1 + ε + β SIZE 4 1 (7) 23
EXFLOW -1 and OFLOW -1 are exchange flow and oher flow during he previous monh respecively. Performance is measured as alpha based on he five performance measures described in he previous secion. The model is execued as a panel regression wih year fixed effecs, wih he sandard errors clusered by fund. Using hese models above o examine performance predicabiliy may cause spurious regression resuls, due o he use of reurns and pas reurns as highlighed by Ferson e al. (2003). As a rule of humb, Ferson e al. saes ha serious spurious regression bias does no arise when he firs order auocorrelaion co-efficien is less han 0.90 and he rue R-square is less han one per cen. All our performance predicabiliy models reurn an auo-correlaion coefficien of less han 0.5 and an adjused R-square of less han one per cen. 7. Resuls 7.1 Flow-Performance Resuls The flow-performance resuls represen our invesigaion ino reurn chasing behavior of he invesors. An observed convex relaionship shows ha invesors have he endency o chase afer high performing funds, while failing o sell ou of he low performing funds. If invesors were informed and raional and he marke mechanism was efficien, we should no observe his relaionship. The resuls from he flow-performance relaionship regression using Morningsar Caegory reurns are shown in Table 4. The resuls for he flow-performance regression using raw Morningsar caegorical rankings yield some resuls which are line wih curren research bu also provide some surprising resuls. Firsly, he unlised consan shows ha an unlised fund experience on average 5.7 basis poins per monh of inflow of funds holding oher variables consan. This reconciles wih ICI s observaion of growh in asse under managemen in he European muual fund indusry from 2002 o 2012. The lised sample however, yields a ne ouflow of 4.4 basis poins per monh on average. Given ha boh coefficiens are significan and consisen across oher performance rankings, i provides evidence ha he exchanges in Germany are used by he invesor as a 24
plaform o dispose of heir muual fund unis. This is a confirmaion of our firs hypohesis: ha lised funds experience a ne ouflow. The use of he exchange channel as a selling avenue exposes he inadequacy of he selling channels in he open-end muual fund indusry. We believe ha hrough a combinaion of managerial incenives and liquidiy issues, here is a rise in he use of he exchange-raded channel as a selling channel. Ineresingly, his finding provides suppor for he Lee-Ready algorihm correcly classifying he rades on he exchanges 9. The use of he exchange channel for disposal of muual fund unis holds imporan implicaions for he muual fund indusry. Firsly, i shows ha he exchange channel benefis he invesor in providing a rading channel, which disrups he sraegic pricing, and sraegies used by funds and brokers o hinder muual fund uni disposal. While his is a good oucome of he exchange-raded channel i comes a he coss of he fund in he form of losing AUM. Expense raio is posiively significan in relaion o flow for boh lised and unlised funds; he effec is sronger for lised funds (31 basis poins) in comparison o unlised funds (19 basis poins). This posiive relaionship beween flow and expense is conrary o empirical sudies by Gruber (1996) and Carhar (1997), which documened a negaive relaionship beween ne expense raio and flow. However, upon furher reading, Barber, Odean and Zheng (2008) found no relaionship beween operaing expenses and flow. However, he auhors did manage o find a srong posiive relaionship beween flow and expenses confined o 12B-1 fees 10. Huang, We and Yan (2007) find expense raio o be posiively relaed o flow afer conrolling for 12B-1 fees and 1/7 fron-end loads. Barber e al. (2008) and Huang e al. (2007) provides suppor for our resuls. A possible explanaion is ha European funds have he endency o spend large amouns on markeing effors o arac flows, causing he relaionship beween expense raio and flow o be posiive. The magniude of he expenses also shows ha lised flow invesors are more sensiive o expenses han unlised fund invesors. Expenses conribue almos 32 basis poins o flow for 9 The Lee-Ready algorihm classifies approximaely 60% of open-end muual fund rades as sells rades and hey are on average larger in magniude han buy rades. 10 Expenses ha are classified for he use of markeing and disribuion. 25
lised funds comparing o 19 basis poins for unlised funds. We believe he high sensiiviy owards fees is driven by he abiliy o reac more quickly o markeing effors by he funds. Boh he magniude and direcion beween flow and expenses indicae ha he exchange-raded channel does no provide any benefis o he invesors, in erms of assising invesors in making more raional decisions. To conclude ha he exchange-raded channel brings benefis o invesors from an expense perspecive, we need o observe a negaive relaionship beween expense and flow. While he resuls and he daa available canno lead o a conclusive explanaion, we believe ha he relaionship warrans furher research and invesigaion 11. Thiry-day reurn volailiy, age, and size for unlised funds, wih he excepion of age are saisically significan and direcionally in line wih exising research (for volailiy and age see Huang, Wei and Yan 2007 and Sialm e al. 2012 and for size see Chen e al. 2004, Yan 2008 and Ferreira e al.2012a. We find Age o have no saisically significan relaionship wih flow. We argue ha age should no have a significan economic impac on flow, given ha he age of a fund ells us nohing regarding he fuure performance of he fund. Oher age facors such as manager enure and board member enure are expeced o have somewha of an effec on performance and flow. While he findings for unlised funds are concurren wih exising empirical evidence; findings for lised flows seem o conradic exising findings. Firsly, we find reurns volailiy no o be a facor deermining flows for lised funds. I is logical and well documened ha muual fund invesors shy away from reurns volailiy (See Hung, Wei and Yan 2007b). The risk adverse naure of invesors and he difficulies in selling ou of muual funds via radiional channels make invesors avoid funds ha exhibi high reurns volailiy. Ineresingly, invesors holding lised funds do no seem o be bohered by reurns volailiy as much as invesors holding unlised funds. This brings back and reinforces he idea ha he exchange channel provides an avenue for which invesors can quickly execue heir rade and pull ou of 11 An aemp was made o source more comprehensive daa on fees and expenses for he fund samples bu due o he differen fee reporing and classificaion rules, we find i difficul o obain usable fee and expense daa for European open-end muual funds. 26
unwaned posiions. So far he exchange channel seems o be providing benefis for invesors mainly hrough speedy execuion of rades. Syle flow is saisically significan for boh lised and unlised funds. This is wihin expecaions, as invesors are likely o exhibi rend chasing behavior as described by Bailey, Kumar and Ng (2011), in paricular for cerain ype of asse classes or syle of funds. For example, ICI records shows here is a srong recen ouflow in equiy funds and a srong inflow ino fixed income/money marke funds, due o he volailiy risk in equiy markes, demonsraing ha general marke rends in flows have an impac on conemporaneous flow. This resul is also empirically suppored by Huang, e al. (2007a) and Sialm e al. (2012). We also noe ha he economic impac of syle flow on unlised funds is much larger han lised funds. This leads us o believe ha unlised fund invesors exhibi much more rend chasing behavior han hose ha rade lised funds. The performance-flow relaionship is highlighed by he coefficiens for low, mid and high 12. Unlised funds, as prediced, produced a classic convex flow performance relaionship as described by Sirri and Tufano (1993 & 1998), Ippolio (1993), Chevalier and Ellison (1997) and Ferrier e al (2012b). Examining he resuls for unlised funds, we also observe ha he relaionship is especially srong for he middle quinile ranked muual funds bu weak for low and high ranked funds. The resuls for he lower quinile however, are saisically weaker han he resuls for middle and op quiniles. The flow-performance sensiiviy is he larges for he highes quinile for unlised funds and lowes for he middle hree quiniles. As for lised funds, he lowes quinile exhibis he highes sensiiviy, relaive o he middle quinile funds. However he middle quinile does no 12 We believe ha he use of caegory percenile ranking (where a fund s raw performance ranking is ranked by oher funds in he same Morningsar caegory and region) is he mos appropriae ranking mehodology. Given our sample is small relaive o oher sudies; a ranking comparison beween funds in he sample may produce inaccurae performance rankings. Benchmark ranking also produced poor resuls across models. We aribue his o he lack of consisenly defined benchmarks wihin he Morningsar daabase; replacemen benchmark is used in he absence of Morningsar defined and fund defined benchmarks. Sensoy (2009) find ha benchmarks defined by US muual funds do no mach he fund s acual syle and is seleced o deliberaely disor a fund s adjused performance o arac inflows. Hence we believe benchmark-adjused reurn is no as effecive as oher adjused reurns. 27
appear o be saisically significan. The highes quinile exhibis slighly less sensiive flowperformance relaionship han he boom quinile. Lised funds also produce a convex flow-performance relaionship as shown by he line wih square markers in Char 2; we noe ha he middle quiniles for lised funds are based on saisically insignifican resuls from he caegorical ranking model above. While he resuls for unlised funds are consisen wih exising lieraure, i.e. producing a convex flow-performance relaionship; resuls for lised funds indicae ha unlised funds have a weakened flow-performance relaionship; especially for flows in he middle performing quinile; conrary o our hird and fourh hypohesis. Lised funds showed srong saisical significance for low and high performing funds using caegorical ranking mehod. This finding does no persis for ranking by oher reurn measures; as ranking by raw reurns yielded significan resuls for he middle quiniles only; while he Carhar, Fama-French, and CAPM adjused reurn ranking yielded significance for he low and middle quinile similar in magniude o ha of he caegorical rankings mehod. The inconsisency of resuls across he differen rankings leads us o believe ha lised open-end muual funds do no exhibi he radiional flow-performance relaionship bu insead he relaionship is weakened, especially for funds ha are in he middle quinile. This finding is concurren wih wha we observe wih syle flow: exchange raded openend muual fund invesors exhibi less reurn chasing behavior, as demonsraed by he lack of saisically and economically significan flow-performance sensiiviy. This provides srong evidence ha invesors who rade lised open-end muual funds exhibi more raional and hence, less rend chasing behavior. This also provides srong evidence for us o rejec our hird hypohesis, which saes ha we should observe a more convex flow-performance relaionship for lised funds. By rejecing our hird hypohesis, i allows us o conclude ha he exchangeraded channel brings benefis o invesors in reducing rend-chasing behavior. However, we mus examine furher evidence o confidenly rejec our hird hypohesis. Unlised funds produce a classic flow-performance relaionship (see line wih diamond marker in Char 2 in he Appendix). The middle and op quinile are he mos saisically 28
significan and his is consisen across Fama-French, and CAPM adjused reurns; which indicaes srong reurn chasing behavior from invesors of unlised funds. Conrary o lised funds, he Carhar-adjused measure does no show any relaionship beween flow and performance for unlised funds. The findings show ha unlised muual fund invesors are momenum-driven and hence, reinforce he view ha hey are reurn chasing. Given he difference in flow-performance behavior for lised and unlised muual funds, we furher invesigae his effec by dividing he flows ino exchange rading and oher (radiional) flows in he nex sage of he analysis. This gives us an in-deph insigh ino how he differen flows inerac wih performance. Bu before ha, we should briefly examine he funcional form of he flow-performance relaionship. Boh lised and unlised funds show a srong linear relaionship beween flow and performance and he findings are consisen across oher performance measures and ranking mehods. However, lised funds exhibi a sligh negaive quadraic funcional form in conras o lised funds, which is observed o have he convenional cubic relaionship as seen in Chavelier and Ellison (1997) and Sialm e al. (2012). The negaive quadraic relaionship for lised funds is saisically srong and consisen across oher performance measures. The negaive quadraic relaionship funcional form gives us more informaion regarding he shape of he flowperformance relaionship for lised funds. The negaive quadraic relaionship indicaes ha he flow-performance relaionship is concave, which suppors our finding in he previous secion, saing an improved raionaliy for lised muual fund invesors. Wih he above resul, we rejec our fourh hypohesis ha he lised muual fund flow-performance relaionship akes he form of a posiive quadraic funcion. Conversely o our consisen findings for lised funds; unlised funds exhibied linear relaionship beween flow and performance consisenly across differen models bu he higher order relaionships are no consisen. CAPM adjused reurns and Fama-French adjused reurns, show a posiive quadraic funcional form, while raw reurn using sample ranking yielded a negaive quadraic relaionship. Carhar - adjused and benchmark - adjused model shows no 29
significan higher order relaionships beween flow and performance for unlised funds. To conclude, we can firmly esablish ha he flow performance relaionship for unlised funds is linear, bu resuls for higher order relaionships are inconsisen and need furher invesigaion. Given he lack of consisency across differen measures, we canno accep nor rejec second he par of our fourh hypohesis. In order o conduc furher invesigaion on flow performance relaionship for lised muual funds, we mus examine flows more closely by using exchange flow and oher flow as our independen variables. To es he robusness of he resuls, we use a more selecive crierion of limiing he op and boom quiniles o en percen. The resuls are presened in Table 5. The heme of ouflow coninues wih he resuls. Exchange flows exhibi unexplained ouflows, alhough he saisical significance is weak. Conrary o resuls from oal flow, age and size boh are economically insignifican and insignifican in deermining flow volume. However, expense provides us wih some ineres resuls. The lack of economic significan resuls shows ha he invesors are no longer concerned wih inefficien funds (hose wih high expense raios) given ha hey are able o sell ou of he fund bypassing he radiional channel. We see his as a significan advanage for he invesor given ha hey are no longer subjec o he sraegic pricing scheme of muual funds. Expense relaes negaively o flow for unlised funds and magniude was amplified o 133 basis poins. This resul is consisen across all oher models. This indicaes ha radiional muual fund flows are exremely sensiive o fund expenses and re-enforces he view ha sraegic fee seing and managerial incenives are highly effecive drivers of muual fund flow. The negaive direcion indicaes ha invesors are using oher channels ha are raional, in ha hey sell ou of funds wih high expense raios. Given ha he findings of he pos-division of flows are significanly differen o findings using oal flow, his validaes he purpose of our paper. As seen by he conrasing resuls, he separaion of flows creaes observaions oherwise no observed previously. This also gives 30
evidence o empirical researchers ha differen channels of flow behave differenly and should be separaed o isolae is effec on performance. The co-efficien for hiry-day reurns volailiy indicaes ha lised muual fund invesors are volailiy-loving. The resuls are economically large a 7.1 and 229 basis poins for exchange and oher flows respecively. The magniude of he co-efficien for exchange flows are in line wih exising lieraure (See Huang e al. 2007 and Sialm e al. 2012). However, oher flows showed a very large posiive sensiiviy o reurns volailiy. This shows ha invesors ha decide o rade hrough radiional channels exhibi gambling-like behavior in order o achieve a higher reurn while ignoring he downside risk. Conrary o our findings from using oal flow as he independen variable, syle flow also becomes negaive, indicaing ha invesors exhibi no rend chasing behavior. This is supporive of he opinion we form afer examining he relaionship beween oal flow and performance. While he resuls are saisically significan, i has lile economic value. Syle flow has 0.02 basis poin and 0.1 basis poin effec on exchange and oher flows respecively. Spliing he flows also reveal a oal breakdown of flow-performance relaionship. Exchange flows exhibi no consisen flow-performance relaionship. Caegories ranking wih raw reurns yield a weak flow-performance relaionship for funds in he middle performance quiniles and oher performance measure yield no significan flow-performance relaionship. This is no consisen wih our hypohesis ha we should observe a more convex flow-performance relaionship for lised and exchange flows. This is a subsanial finding in ha he flowperformance relaionship is so well esablished in exising lieraure. Oher flows demonsrae weak evidence of flow-performance relaionship for he boom and middle quiniles, bu he resuls are only found using raw caegorical rankings, Fama- French and Carhar-adjused performance. CAPM adjused performance yield a significan relaionship for funds ha perform in he highes quinile. Boh benchmark and raw reurn rankings have no saisically significan relaionship beween flow and performance. 31
We re-ierae he imporance of our findings and he need o separae channels of flows; flow-performance relaionship is empirically observed in many foundaion papers in muual fund research (Gruber 1996, Sirri and Tufano 1993 & 1998, Chevalier and Ellison 1997 and Del Guercio and Tkac 2002), his sudy essenially challenges he findings of hose papers. The resuls also highligh he imporance no o assume flow behavior is he same across differen channels. 7.3 Performance-Predicabiliy Resuls Zheng (1999) s seminal paper on smar muual fund flows spawned a hos of muual fund sudies (see Sapp and Tiwari 1994, Frazzini and Lamon 2008, for example) around he performance predicabiliy of muual fund flows. Following on from our sudy on flowperformance relaionship, we also furher invesigae he predicive abiliy of flows. Lised flows are found o be consisenly smar across he differen performance measures wih he excepion of Carhar-Adjused measure. The disappearance of he smar money effec afer conrolling for momenum is documened by Sapp and Tiwari (2004). For unlised funds however, we find no evidence of smar money effec, wih he excepion of using he benchmark-adjused performance measure. Our resuls for unlised funds are differen o hose of Zheng (1999). We aribue his o he difference in he fund regions of our sample. While Zheng (1999) solely focused on US Equiy funds, our sample consiss of a mixure of funds from across he European Union and asse classes. We see his as a poenial source of where he discrepancy arises. Lised funds, however, yield a much more ineresing resul. When we pariion oal flow ino exchange and oher flows, we see a significan smarness of exchange flow and a relaive dumbness of flow from oher channels. Exchange flows show performance predicabiliy across differen reurn measures, wih he excepion of caegory-adjused reurns. This confirms our fifh hypohesis: exchange flows are smarer han radiional flows. This does no come as a surprise, given ha mos invesors using hese exchange raded plaforms are more sophisicaed, relaive o invesors using more radiional channels. 32
Oher flows exhibied a dumb money effec as described in Frazzini and Lamon (2008). Frazzini and Lamon concludes ha individual invesors rade poorly and ha heir flows exhibi a very weak smar money effec in he shor erm and are dumb in he long erm. We apply a similar concep in explaining our observaion. We believe ha more sophisicaed invesors use hese exchange plaforms o quickly execue heir rading sraegies while he nonsophisicaed invesors who use he oher flow channels lag behind and are unable o correcly idenify managerial alen and hence, causing he non-exchange raded channels o be dumb. Wermers (2004), Coval and Safford (2007) hypohesized ha muual fund flows increase fund prices, which drives he smar money effec, while Sapp and Tiwari (2004) ried o explain he smar money effec using momenum effecs. These are all valid empirical explanaions for smar money flow. However in our case, we believe ha no maer wha he asse class, informaion asymmery exiss beween differen invesors. There is no doub ha here are informed and uninformed invesors in he European open-end muual fund space. We believe ha hese more informed invesors are likely o use he exchange-rading plaform o execue heir sraegies. While his benefis he informed rader, i comes a he expense of he uninformed invesors in he marke. 7.3 Invesor Sophisicaion While he purpose of his sudy is no o explore he relaionship beween invesor sophisicaion and channel choice; we can possibly make some preliminary observaions regarding wheher he exchange-raded channel is used by more sophisicaed invesors and wheher he exchange-channel allows us o observe he rading skills of more sophisicaed invesors. To conclude our discussion, we would like o make some preliminary links beween invesor sophisicaion and he exchange-raded channel, based on exising lieraure and evidence from our resuls. We mus noe ha wihou he availabiliy of clearing house daa, we canno draw sric conclusions regarding he ype of invesors rading via he exchange-raded channel. 33
Hackehal e al. (2012) using German brokerage daa, found ha invesors wih advisors feaure higher urnovers bu earn lower ne reurns, as well as lower Sharpe raios. These invesmen advisors will purchase and sell ou of muual fund unis direcly wih he muual funds via he radiional channel 13. This leads us o believe ha sophisicaed and informed invesors will no rade wih an accoun aached o an advisor bu insead will rade via an online broker. Hence, we reason ha more sophisicaed and more informed raders mainly use he exchangeraded channels, wheher hey are reail or insiuional. This holds an imporan implicaion, if our reasoning was proved correc, hen empirical researchers would be able o make observaions regarding sophisicaed invesors by exclusively examining he exchange raded channel. The above reasoning also provides suppor and an explanaion for our fifh hypohesis: ha exchange flows are smarer. Flows ha originae from he exchange are smarer, reconciles wih our reasoning above. We believe hrough our reasoning we can infer ha informed and sophisicaed invesors are he dominan users of he exchange raded channel, causing he exchange flow o exhibi smar properies, i.e. he abiliy o predic fuure performance. 7.4 Robusness Checks Following he lieraure (Sirri and Tufano 1998 and Huang e al. 2007a) we acknowledge ha flows may be affeced by he ineracion of fund reurns and non-performance relaed fund characerisics. As an added robusness check we conduc a series of regressions incorporaing ineracion erms beween reurns and oher fund characerisics. Firsly, a regression wih an ineracion beween expenses and he hree ranking quinile variables (Low, Mid, High) along wih an age and performance ineracion erm o conrol for he poenial ineracion effecs beween expense and performance as well as age and performance. 13 Afer consuling wih muual fund advisors and researching ino he dynamics of he muual fund indusry, we find ha muual funds will disribue fund unis hrough he commission driven channel, i.e. hrough advisors a financial inermediaries. Funds used a variey of mehods o arge hese advisors o disribue heir unis; some examples include offering discouns on bulk allocaions, disribuion agreemens and commission based on sales volumes. Hence we believe ha he advisors have no incenive o sell ou or buy in o muual fund unis via he exchange channel. 34
A second se of ineracion uilizes a dummy which indicaes one if he chosen variables (age, size and expense) is greaer han he median wih he hree ranking variables (Low, Mid, High). This gives us o he abiliy examine wheher flow-performance relaionship is driven by funds ha are older, larger or hose ha have higher expenses. This robusness check allows us o make generalized conclusions regarding our daa se, no jus specific ypes of funds. The resuls for he dummy ineracion erms are mosly insignifican wih he excepion of middle performing ranking funds using he raw-reurn measure, as well as for lised funds using he Carhar four-facor model ranking mehod. The lack of consisency and significance across differen performance rankings indicae o us ha our resuls are no driven by only larger, older or more expense-inensive funds, hence our discussion in he previous secions are sill valid. 14 14 A furher aemp is made o examine he effecs of he marke maker on our resuls and conclusion. We divide he exising samples for our ess ino wo samples, one before he inroducion of he marke maker and one pos. Afer running our analysis on he wo samples we come o he conclusion ha here are insufficien observaions in he sample represening he relaionship prior o he inroducion of he marke maker. The sample only represens less han 9% of he oal sample of observaions. The resuls for he pos-marke maker sample are similar o he resuls already obained. 35
8. Conclusion Our sudy of wo samples of muual funds, one wih lising channels and one wihou, provide subsanial resuls beween flow and performance. Firsly, we find srong evidence of ne ouflow for lised muual funds while unlised funds experience ne inflow of funds. This srong evidence of ne ouflow suggess ha he exchange-raded channels faciliae he invesor s need o sell ou of muual fund posiions bypassing he slow and expensive channels. Afer examining he flow and performance of our samples, we found evidence for flowperformance relaionship as described by Sirri and Tufano (1998) for unlised funds; however he relaionship is significanly weakened for lised muual funds, which indicaes ha lised muual fund invesors exhibi less rend chasing behavior. We see his as a significan benefi o he invesors in he form of reduced reurn chasing behavior due o he flexibiliy o buy in and sell ou of fund posiions. Boh lised and unlised funds exhibi linear flow-performance funcional form. Unlised funds have a sligh posiive cubic funcional as found by Sialm e al. (2012) and lised funds were found o have a sligh negaive quadraic relaionship funcional form. Reversing he flow-performance relaionship ess, we conduc a performance predicabiliy es similar o ha of Zheng (1999) and Sialm e al. (2012) o es he performance predicabiliy of flows. Lised fund flows are found o have significan predicive abiliy in comparison o unlised funds, which do no show evidence of he smar money effec. Furher o his, we find ha funds ha originae from he exchange are smar and funds from he radiional channels are dumb, i.e. does no offer any predicive abiliy. While hese resuls are significan, i has lile impac due o he small magniude of he coefficiens. However, we can conclude ha he predicive abiliy of lised flows originaes from he exchange channel. In conclusion, we believe ha more sophisicaed and informed invesors prefer o rade via he exchange-raded channels. As a resul of his, we observe boh a breakdown in he flowperformance relaionship and exchange flows showing performance predicabiliy. 36
The inroducion of exchange rading for open-end funds raises concerns abou he coss of higher frequency rading in he same funds ha are sill open o rading hrough radiional channels. The lae rading scandal of he mid 2000's as highlighed by Greene and Hodges (2002) and Zizewiz (2003) describes how muual fund raders were able o earn abnormal reurns hrough exploiing he mismach in close of day NAV calculaion and any rades conduced pos he pricing period. This loophole gives a large and highly acive rader he abiliy o earn arbirage reurns a he expense of he passive shareholders. We believe a naural exension of his paper is o examine wheher he exchange raded channel will ameliorae he effec oulined by Greene e al. and Zizewiz, i.e. wheher or no he exchange raded channel can be used as a mechanism o reduce abnormal reurns for acive and large raders and bring benefis o passive muual fund holders. Given our sudy indicaes ha expenses play a srong role in he relaionship beween flow and performance, anoher exension would be o look a he impac of he exchange raded channel on fee srucures. Tradiionally, fee arrangemens such as back end loads are designed o ameliorae concerns abou he coss of frequen wihdrawals on long-erm fund invesors. However, wih he inroducion of he exchange rading plaforms, invesors may bypass hese loads and disrup he sraegic pricing schemes of he fund. Anoher possible exension o his paper is o explore he marke design and marke microsrucure of he exchange raded muual fund marke wih a focus on he role of he marke maker. Sudy by Reiss and Werner (2004) found ha marke archiecure and microsrucure significanly affecs he rader s choice of rading channels. This exension will formally es he informaion conained in he rades from he exchanges and furher invesigae he idiosyncrasies of flows ha originae from he muual fund exchanges. Ulimaely, he exchange-raded channel does bring benefis o invesors, hough a he expense of he open-end muual fund indusry. The exchange-raded channel faciliaes he need for invesors o sell ou and buy ino posiions in a speedy manner, bypassing he radiional channels ha are deemed oo slow and inefficien. While his seems o aid he ouflow of money 37
from he indusry, he exchange-raded channel uninenionally serves as a monioring device for open-end muual fund invesors. While his does no seem o benefi he indusry a firs, i serves as a reminder ha invesors now have he means o sell ou of poor performing funds more freely and perhaps can improve he governance and compeiiveness of he indusry in he fuure. We see his as paricularly imporan, given he rise in alernaive invesmen vehicles for invesors, mainly in he form of ETFs. The resuls of his paper also carry policy implicaions. As menioned previously, auhoriies in he pos-global financial crisis world are increasingly pushing for more ransparency for asses and accounabiliy of invesmen managers. We have shown in our research ha public lising does bring benefis o he invesor, by giving hem he means o sell ou of poor performing asses and mos imporanly, he exchange can be used as an informal monioring device for invesmen managers. 38
Appendix A Char 1 Ne Muual Fund Flows via German Exchanges Char 1 is year-by-year nominal amoun of flow (measured by muliplying volume and price and aking he sum across differen years) across all six German exchanges wih open-end muual fund rading. Dusseldorf, Berlin, Frankfur, Sugar, Hamburg and Munich. 2012 volume only includes flow up o June, 2012. All yearly flows are measured in millions of Euros. 500 0-500 -1,000-1,500-2,000 1996 1998 2000 2002 2004 2006 2008 2010 2012 Berlin Dusseldorf Frankfur Hamberg Munich Sugar Grand Toal 39
Char 2 Flow-Performance Relaionship for Lised and Unlised Funds Char 2 shows he resul of he flow-performance relaionship for lised and unlised funds using ranking by raw reurns from Morningsar. The y-axis represens he percenage of he fund s TNA ha are raded, and he x-axis represens he percenile he fund is ranked relaive o is peers in he same Morningsar Caegorical ranking. Noe ha a rank of 100 represens he op performing funds while 0 represens he lowes performing funds. 2.5 2 1.5 1 0.5 0 0 20 40 60 80 100 LISTED UNLISTED 40
Appendix B 1. Definiion of Variables Variable Ln(Age) Ln(Size) Expense Syle Flow Toal Exflow Oflow Vol. Low Mid High MC_Low MC_Mid MC_High 4FF_Low 4FF_Mid 4FF_High Age*Perf Low,Mid,High_Exp Low,Mid,High_Exp_D Low,Mid,High_Size_D Low,Mid,High_Age_D Definiion Logarihm of he number of days since he fund s incepion Logarihm of he capializaion of he fund (calculaed as number of shares ousanding by price) The annual ne expense raio for he fund expressed as a percenage of asse under managemen The average percenage of monhly flow ino cerain Morningsar caegory The oal monhly percenage flow ino Fund I The amoun of percenage ha is raded via public exchanges Toal amoun of flow for a fund minus any exchange flow Monhly reurn volailiy, calculaed as he hiry day sandard deviaion of reurns Amoun of percenile conribuion of he lowes quinile o he fund s raw reurn percenile ranking relaive o peers in he same sample. Amoun of percenile conribuion of he middle hree quiniles o he fund s raw reurn percenile ranking relaive o peers in he same sample. Amoun of percenile conribuion of he highes quinile o he fund s raw reurn percenile ranking relaive o peers in he same sample. Amoun of percenile conribuion of he lowes quinile o he fund s raw reurn percenile ranking relaive o peers in he same Morningsar Caegory. Amoun of percenile conribuion of he middle hree quiniles o he fund s percenile ranking relaive o peers in he same Morningsar Caegory. Amoun of percenile conribuion of he highes quinile o he fund s percenile ranking relaive o peers in he same Morningsar Caegory. Amoun of percenile conribuion of he lowes quinile o he fund s four facor adjused reurn percenile ranking relaive o peers in he same sample. Amoun of percenile conribuion of he middle hree quiniles o he fund s four facor adjused reurn percenile ranking relaive o peers in he same sample. Amoun of percenile conribuion of he highes quinile o he fund s four facor adjused reurn percenile ranking relaive o peers in he same sample. Ineracion erm beween age and performance Ineracion erm beween quinile variables low, mid and high wih expense Ineracion erm beween quinile variables low, mid and high wih expense dummy which indicaes 1 if he expense is above he sample median Ineracion erm beween quinile variables low, mid and high wih size dummy which indicaes 1 if he expense is above he sample median Ineracion erm beween quinile variables low, mid and high wih age dummy which indicaes 1 if he expense is above he sample median 41
Figure 1: Visual represenaion of differen channels in European Exchange Traded Open-End Muual Funds marke 42
References Adams, J., Mans S., & Nishikawa, T. "Are Muual Fund Fees Excessive." Journal of Banking and Finance, 2012: 2245-2259. Ammann, M., Ising, A., & Kessler, S. "Disposiion Effec and Muual Fund Performance." Unpublished Manual Scrip, 2011. Aiken, M., Frino, A. "The Accuracy of he Tick Tes: Evidence from he Ausralian Sock Exchange." Journal of Banking and Finance, 1996: 1751-1729. Bailey, W., Kumar, A., & Ng, D. "Behavioural Biases of Muual Fund Invesors." Journal of Financial Economics, 2011: 1-27. Barber, B., Odean, T., & Zheng, L. "Ou of Sigh, Ou of Mind: The Effecs of Expenses on Muual Fund Flows." Journal of Business, 2005: 2095-2120. Bergsresser, D., Chalmers, J., & Tufano, P. "Assessing he Coss and Benefis of Brokers in he Muual Fund Indusry." Review of Financial Sudies, 2009: 4129-4156. Carhar, M. Carpener, J., Lynch, A., & Muso, D. "Muual Fund Survivorship." Review of Financial Sudies, 2002: 1439-1463. Carhar, M. "On Persisence in Muual Fund Performance." Journal of Finance, 1997: 57-82. Chalers, J., Edelen, R., & Kadlec, G. "On he Perils of Securiy Pricing by Financial Inermediaries: The Case of Open-End Muual Funds." The Journal of Finance, 2001: 2009-2236. Chen, J., Hong, H., Huang, M., & Kubik, J. "Does Fund Size Erode Muual Fund Performance? The Role of Liquidiy and Organizaion." American Economic Review, 2004: 1276-1302. Chevalier, J., & Ellison, G. "Risk Taking by Muual Funds as a Response o Incenives." Journal of Poliical Economy, 1997: 1167-1200. Chordia, T. "The Srucure of Muual Fund Charges." Journal of Financial Economics, 1996: 3-39. Cic G. "The Relaion of he Disposiion Effec o Muual Fund Trades and Performance." Unpublished Manual Scrip, 2010. Coval, J., & Safford, E. "Asse Fire Sales (and Purchases) in Equiy Markes." Journal of Financial Economics, 2007: 479-512. Del Guercio, D., & Tkac, P. "The Deerminans of he Flow of Funds of Managed Porfolios: Muual Funds vs. Pension Funds." Journal of Financial and Quaniaive Analysis, 2002: 523-557. Ellis, K., Michaely, R., & O'Hara, M. "The Accuracy of Trade Classificaion Rules: Evidence from NASDAQ." Journal of Financial and Quaniaive Analysis, 2000: 529-551. Elon, J., Gruber, M., & Blake, C. "A Firs Look a he Accuracy of he CRSP Muual Fund Daabse and a Comparison of he CRSP and Morningsar Muual Fund Daabase." Journal of Finance, 2001: 2415-2430. Fama, E. "Efficien Capial Markes: A Review of Theory and Empirical Work." Journal of Finance, 1970: 388-417. 43
Fama, E., & French, K. "Common Risk Facors in he Reurns on Socks and Bonds." Journal of Financial Economics, 1993: 3-56. Fama, E., & Macbeh, J. "Risk, Reurn, and Equilibirum Tess." Journal of Poliical Economy, 1973: 607-636. Ferreira, M., Keswan A., Miguel, A., & Ramos, S. "The Deerminans of Muual Fund Performance: A Cross-Counry Sudy." Review of Finance, 2012: 1-43. Ferreira, M., Keswan A., Miguel, A., & Ramos, S. "The Flow-Performance Relaionship Around he World." Journal of Banking and Finance, 2012: 1759-1780. Ferson, W., Sarkissian, S., & Simin, T. "Spurious Regressions in Financial Economics." Journal of Finance, 2003: 1393-1414. Frazzin A., & Lamon, O. "Dumb Money: Muual fund flows and he cross-secion." Journal of Financial Economics, 2008: 299-322. Gil-Bazo, J., & Ruiz-Verdu, P. "The Relaion Beween Price and Performance in he Muual Fund Indusry." Journal of Finance, 2009: 2153-2183. Goezmann, W., & Peles, N. "Cogniive Dissonance and Muual Fund Invesors." Journal of Financial Research, 1997: 145-158. Greene, J., & Hodges, C. "The Diluion Impac of Daily Fund Flows on Open-End Muual Fund." The Journal of Financial Economics, 2001: 131-158. Grinbla, M, and S & Timan. "The Persisence of Muual Fund Persisence." Journal of Finance, 1992: 1977-1984. Grinbla, M., & Timan, S. "A Sudy of Monhly Muual Fund Reurns and Porfolio Performance Evaluaion Techniques." Journal of Financial and Quaniaive Analysis, 1994: 419-444. Gruber, M. "Anoher Puzzle: The Growh in Acively Managed Muual Funds." The Journal of Finance, 1996: 783-810. Hackehal, A., Haliassos, M., & Jappell T. "Financial Advisors: A Case of Babysiers?" Journal of Banking and Finance, 2012: 509-524. Hendricks, D., Pael, J., & Zeckhauser, R. "Ho Hands in Muual Funds: Shor-Run Persisence of Relaive Performance." Journal of Finance, 1993: 93-130. Huang, J., We K., & Yan, H. "Paricipaion Coss and he Sensiiviy of Fund Flows o Pas Performance." Journal of Finance, 2007: 1273-1311. Huang, J., We K., & Yan, H. "Volailiy of Performance and Muual Fund Flows." Universiy of Texas Working Paper, 2007. "Invesmen Company Fac Book." 2011. hp://www.icifacbook.org/ (accessed 2012). Ippolio, R. "Consumer Reacion o Measures of Poor Qualiy: Evidence from he Muual Fund Indusry." Journal of Law and Economics, 1992: 45-70. 44
Ivkovic, Z., & Weisbenner, S. "Individual Invesor Muual Fund Flows." Journal of Financial Economics, 2009: 223-237. Jegadeesh, N., and S. & Timan. "Reurns o Buying Winners and Selling Losers: Implicaions for Sock Marke Efficiency." Journal of Finance, 1993: 65-91. Kahneman, D., & Tversky, A. "Prospec Theory: An Analysis of Decision Under Risk." Economerica, 1979: 263-292. Kaplan, S., & Schoar, A. "Privae Equiy Performance: Reurns, Persisence, and Capial Flows." Journal of Finance, 2005: 1791-1823. Keswan A., & Solin, D. "Which Money is Smar? Muual Fund Buys and Sells of Individual and Insiuional Invesors." Journal of Finance, 2008: 85-118. Khorana, A., Servaes, H., & Tufano, P. "Explaining he Size of he Muual Fund Indusry Around he World." Journal of Financial Economics, 2005: 145-185. Khorana, A., Servaes, H., & Tufano, P. "Muual Fund Fees Around he World." Review of Financial Sudies, 2009: 1279-1310. Lee, C., & Radharkrishna, B. "Inferring Invesor Behavior: Evidence from TORQ daa." Journal of Financial Markes, 2000: 83-111. Lee, C., & Ready, M. "Inferring Trade Direcion from Inraday Daa." Journal of Finance, 1991: 733-747. Machina, M. ""Expeced Uiliy" Analysis wihou he Indepedence Axiom." Economerica, 1982: 277-323. Odean, T. "Are Invesors Relucan o Realize Their Losses?" Journal of Finance, 1998: 1775-1798. Oen, R., & Bams, D. "European Muual Fund Performance." European Financial Managemen, 2002: 75-101. Reiss, P., & Wener, I. "Anonymiy, Adverse Selecion, and he Soring of Inerdealer Trades." Review of Financial Sudies, 2004: 600-636. Rosenberg, J., & Traub, L. "Price Discovery in he Foreign Currency Fuures and Spo Marke." Federal Reserve Bank of New York Repor no.262, 2006. Rouwenhors, K. "The Origins of Muual Funds." Yale School of Managemen Working Paper, 2004. Sapp, Travis., & Tiwar A. "Does Sock Reurn Momenum Explain he "Smar Money" Effec." Journal of Finance, 2004: 2605-2622. Sensoy, B. "Performance Evaluaion and Self-Designaed Benchmark Indexes in he Muual Fund Indusry." Journal of Financial Economics, 2009: 25-39. Shefrin, H., & Saman, M. "The Disposiion o Sell Winners Too Early and Ride Looser Too Long: Theory and Evidence." Journal of Finance, 1985: 777-790. Sialm, C., Sarks, L., & Zhang, H. "Defined Conribuion Pension Plans: Sicky or Discerning Money?" Universiy of Texas Working Paper, 2012. 45
Singal, V., & Xu, Z. "Selling Winners, Holdings Losers: Effec on Fund Flows and Survival of Disposiion-Prone Muual Funds." Journal of Banking and Finance, 2011: 2704-2718. Sirr E., & Tufano, P. "Cosly Search and Muual Fund Flow." The Journal of Finance, 1998: 1590-1622. Sirr E., & Tufano, P. "Compeiion and Change in he Muual Fund Indusry." Financial Services:Perspecive and Challenges, 1993. Swizer, L., & Fan, H. "Ineracions Beween Exchange Traded Derivaives and OTC Derivaives: Evidence for he Canadian Dollar Fuures vs. OTC Markes." Inernaional Journal of Business, 2008: 25-42. Wermers, R. "Is money really smar? New evidence on he relaion beween muual fund flows, manager behavior, and performance." Working Paper Universiy of Maryland., 2004. Yan, X. "Liquidiy, Invesmen Syle, and he Relaion beween Fund Size and Fund Performance." Journal of Financial and Quaniaive Analysis, 2008: 741-768. Zheng, L. "Is Money Smar? A Sudy of Muual Fund Invesors' Fund Selecion Abiliy." The Journal of Finance, 1999: 901-933. Zizewiz, E. "How Widespread Was lae Trading in Muual Funds?" American Economic Review, 2006: 284-289. 46
Table 1 Domicile Breakdown of Lised and Unlised Sample Table 1 is a crossed secional represenaion of he coun, size and year of lised and unlised sample of funds as of June, 2012. Size is measured as he TNA of he fund a he end of June 2012, and age of he fund is measured as he number of years beween he end of June 2012 and he fund s incepion dae. Domicile No. Size ( m)* Age(Year)* No. Size ( m)* Age(Year)* Ausria 19 272.48 15.52 113 347.27 11.60 Germany 227 677.19 21.05 47 788.48 4.43 France 10 1,277.98 21.45 182 679.83 10.58 UK 15 1,857.71 20.74 69 1,464.65 8.56 * As of June, 2012 Lised Unlised 47
Table 2 Resul of Sample Mean T-es Table 2 consiss of resuls of our -es for difference in mean across our wo samples. The firs column represens he combined average of all values in sample, he second column represens he co-efficien of he -es for differences in mean, and he las column represens he p-value of he null hypohesis which saes ha he difference of he wo samples amouns o zero. Hence, if he p-value is no zero hen we can conclude ha he variable for lised and unlised funds is no differen from 0. The sandard deviaion of he variable is saed in he parenheses. Significance levels are saed a 10%, 5% and 1% using *, **, and *** respecively. Variable Combined Average Difference H0: Diff = 0 Age (Days) Size ( b) Expense (%) Volailiy Syle Flow (%) Reurn (%) FF_RET TF_RET CAPM_RET 4340.8411-2849.0392*** (28.2411) (28.2411) 0.7692-0.0300** (0.0068) (0.0150) 1.4124 0.0382**** (0.0039) (0.00795) 3.3686-1.0882*** (0.0104) (0.0205) 8.1794 12.89342*** (0.7740) (1.5459) 0.4485-0.0733** (0.0167) (0.0334) -0.0183 0.0346 (0.0122) (0.0244) 0.0134 0.0146 (0.0129) (0.0258) 0.0454-0.0166 (0.0135) (0.2701) 0.0000 0.0458 0.0000 0.0000 0.0000 0.0283 0.1574 0.5719 0.5383 48
Table 3 Buy-Sell Breakdown for Open-end Muual Funds from all Six Germany Exchanges Table 3 shows a summary of saisics of he buy and sell direcions of open-end muual funds from our sudy period of January 2002 o June, 2012. The rade direcions are deermined by he Lee-Ready Algorihm (See Lee and Ready 1991). Exchange Buy Sell % BUY % SELL Dusseldorf 103,453 111,428 48.14% 51.86% Berlin 137,238 168,489 44.89% 55.11% Farnkfur 413,940 561,010 42.46% 57.54% Sugar 139,871 217,551 39.13% 60.87% Hamburg 516,922 515,204 50.08% 49.92% Munich 75,566 123,370 37.99% 62.01% 49
Table 4 Flow-Performance Relaionship: Lised Vs. Unlised Funds Table 4 conains he coefficiens of piecewise ime-fixed effec panel regression of he lised and unlised sample o es he flow-performance relaionship. The dependen variable is oal fund flow, which is calculaed as he excess percenage growh of a fund ha would have occurred naurally. ln(age) is he logarihm of he fund s age, which is measured as he number of days since incepion and monh -1. ln(size) is he logarihm of he fund s size, expense is he ne expense raio of he fund for monh -1, 30 day volailiy is he sandard deviaion of daily reurns during he monh -1, and syle flow is he percenage flow ino fund i s Morningsar caegory. The performance componen is represened by Low, Mid, and High variables; where Low, Mid and High represen he percenile rank of he fund s raw performance agains peers in he same Morningsar caegory and region; specifically, he hree performance quiniles are calculaed by he following Low f, = min(rank f,, 0.2), Mid f, = min(rank f, - Low f,, 0.2), and High f, = Rank f, - Low f, Mid f,.. Difference represens he difference beween he lised and unlised co-efficien. The sandard errors clusered by fund are saed in he parenheses. Significance levels are saed a 10%, 5% and 1% using *, **, and *** respecively. Variables Lised Unlised Difference Consan ln(age) ln(size) Expense 30 Day Volailiy Syle Flow Low Mid High -0.0464*** 0.0570*** 0.1095*** (0.0099) (0.0152) (0.0176) 0.0013-0.006*** -0.0073*** (0.0008) (0.0011) (0.0014) 0.0007** -0.001-0.0018** (0.0004) (0.0006) (0.0007) 0.2747** 0.1935*** -0.1290 (0.1096) (0.0651) (0.1462) -0.0854-0.4661*** -0.3338** (0.0958) (0.0943) (0.1407) 0.0013** 0.2315*** 0.2284*** (0.0007) (0.066) (0.0608) 0.0447*** 0.0300* -0.0287 (0.0149) (0.0172) (0.0178) 0.0049 0.0092*** -0.0015 (0.0033) (0.0036) (0.0088) 0.0369*** 0.0362** 0.0584* (0.0139) (0.0166) (0.0352) Observaions 4,880 6,464 11,343 R-Square 0.0101 0.0180 0.0163 50
Table 5 Flow-Performance Relaionship: Lised Vs. Unlised Funds Using Difference Performance Measures Table 5 conains he coefficiens of piecewise ime-fixed effec panel regression of he lised and unlised sample o es he flow-performance relaionship. The dependen variable is oal fund flow, which is calculaed as he excess percenage growh of a fund ha would have occurred naurally. ln(age) is he logarihm of he fund s age, which is measured as he number of days since incepion and monh -1. ln(size) is he logarihm of he fund s size, expense is he ne expense raio of he fund for monh -1, 30 day volailiy is he sandard deviaion of daily reurns during he monh -1, and syle flow is he percenage flow ino fund i s Morningsar caegory. The performance componen is represened by Low, Mid, and High variables; where Low, Mid and High represen he percenile rank of he fund s various performance measures agains peers in he same sample; specifically, he hree performance quiniles are calculaed by he following Low f, = min(rank f,, 0.2), Mid f, = min(rank f, - Low f,, 0.2), and High f, = Rank f, - Low f, Mid f,. The sandard errors clusered by fund are saed in he parenheses. Significance levels are saed a 10%, 5% and 1% using *, **, and *** respecively. Variables Lised Unlised Lised Unlised Lised Unlised Lised Unlised Lised Unlised Consan ln(age) ln(size) Expense 30 Day Volailiy Syle Flow Low Mid Raw Reurn Carhar - Adjused Fama French- Adjused CAPM - Adjused Benchmark - Adjused -0.0453*** 0.0557*** -0.0489*** 0.0681*** -0.0465*** 0.0632*** -0.0499*** 0.0609*** -0.0374*** 0.0679*** 0.0100 (0.0153) (0.0099) (0.0151) (0.0099) 0.0149 (0.0099) 0.015 (0.0099) 0.0152 0.0012-0.0061*** 0.001-0.006*** 0.0011-0.006*** 0.0012-0.0061*** 0.0013-0.0059*** (0.0008) (0.0011) (0.0008) (0.0011) (0.0008) 0.0011 (0.0008) 0.0011 (0.0008) 0.0011 0.0007** -0.0011* 0.0007* -0.0012* 0.0007** -0.001* 0.0007* -0.0009 0.0009** -0.0012* (0.0004) (0.0006) (0.0004) (0.0006) (0.0004) 0.0006 (0.0004) 0.0006 (0.0004) 0.0006 0.2926*** 0.1654*** 0.4636*** 0.184*** 0.4275*** 0.1608** 0.4549*** 0.1744*** 0.3033*** 0.1911*** (0.1092) (0.0643) (0.1088) (0.0648) (0.1092) 0.0653 (0.109) 0.0652 (0.1098) 0.0658-0.1226-0.3766*** -0.0764-0.5235*** -0.0953-0.5387*** -0.0688-0.541*** -0.137-0.5204*** (0.0945) (0.0919) (0.1002) (0.0922) (0.1023) 0.0952 (0.1037) 0.0973 (0.0959) 0.0929 0.0013* 0.2196*** 0.0019*** 0.231*** 0.0017** 0.2143*** 0.0016** 0.2066*** 0.0012* 0.2083*** (0.0007) (0.0655) (0.0007) (0.0652) (0.0007) 0.0652 (0.0007) 0.0652 (0.0007) 0.0666 0.0545*** 0.0414*** 0.0513*** 0.0016 0.0431*** 0.0015 0.0591*** 0.0016-0.0072-0.0013 (0.0159) (0.0159) (0.0141) (0.0151) (0.0143) 0.0144 (0.0142) 0.0145 (0.0141) 0.0154 0.0005 0.0139*** 0.0155*** -0.0007 0.0085*** 0.0091*** 0.0081** 0.012*** -0.0003 0.0053 (0.0032) (0.0036) (0.0032) (0.0036) (0.0032) 0.0034 (0.0032) 0.0034 (0.0033) 0.0037 0.014-0.0058-0.0216 0.0295* -0.0075 0.0592*** -0.0115 0.0473*** 0.0172-0.0108
Table 6 Flow-Performance Funcional Form: Lised Vs. Unlised Funds Table 6 conains he coefficiens of piecewise ime-fixed effec panel regression of he lised a sample o es he funcional form of he flow-performance relaionship. The dependen varia flow, which is calculaed as he excess percenage growh of a fund ha would have occurred ln(age) is he logarihm of he fund s age, which is measured as he number of days since inc monh -1. ln(size) is he logarihm of he fund s size, expense is he ne expense raio of he f -1, 30 day volailiy is he sandard deviaion of daily reurns during he monh -1, and syle f percenage flow ino fund i s Morningsar caegory. The funcional performance componen by Linear, Quadraic, and Cubic variables; where Linear, Quadraic and Cubic represen he f of he flow-performance relaionship of he fund s raw performance agains peers in he sam caegory and region; specifically, he hree funcional forms are calculaed by he following Li min(rank f, 0.5), Quadraic f, = min(rank f, 0.5) 2, and Cubic f, = (Rank f, 0.5) 3. The sanda clusered by fund are saed in he parenheses. Significance levels are saed a 10%, 5% and and *** respecively. LISTED UNLISTED Variables Linear Non - Linear Linear Non - Li Consan ln(age) ln(size) -0.0363*** -0.0352*** 0.066*** 0.0651* (0.0097) (0.0097) (0.0151) (0.0151 0.0013 0.0013-0.006*** -0.006* (0.0008) (0.0008) (0.0011) (0.0011 0.0008** 0.0008** -0.0011* -0.001 (0.0004) (0.0004) (0.0006) (0.0006 Expense 30 Day Volailiy Syle Flow Linear 0.3215*** 0.3271*** 0.198*** 0.1904* (0.1088) (0.1089) (0.065) (0.0652-0.1432-0.1305-0.4677*** -0.4659* (0.0946) (0.096) (0.0944) (0.0944 0.0014** 0.0014** 0.2337*** 0.2298* (0.0007) (0.0007) (0.066) (0.0661 0.0092*** 0.0069 0.0136*** 0.0051 (0.0019) (0.0045) (0.0022) (0.0053 Quadraic Cubic - -0.0138* - 0.0059 - (0.0074) - (0.0087-0.0171-0.0605 - (0.0285) - (0.0348 Observaions 4,880 4,880 6,610 6,610 R-Square 0.0094 0.0102 0.0176 0.0181 52
Table 7 Flow-Performance Funcional Form for Lised Funds Table 7 conains he coefficiens of piecewise ime-fixed effec panel regression of he lised sample o es he funcional form of he flow-performance relaionship. The dependen variable is oal fund flow, which is calculaed as he excess percenage growh of a fund ha would have occurred naurally. ln(age) is he logarihm of he fund s age, which is measured as he number of days since incepion and monh -1. ln(size) is he logarihm of he fund s size, expense is he ne expense raio of he fund for monh -1, 30 day volailiy is he sandard deviaion of daily reurns during he monh -1, and syle flow is he percenage flow ino fund i s Morningsar caegory. The funcional performance componen is represened by Linear, Quadraic, and Cubic variables; where Linear, Quadraic and Cubic represen he funcional form of he flowperformance relaionship of he fund s performance specified in he firs row of he able agains oher funds in he sample; specifically, he hree funcional forms are calculaed by he following Linear f, = min(rank f, 0.5), Quadraic f, = min(rank f, 0.5) 2, and Cubic f, = (Rank f, 0.5) 3. The sandard errors clusered by fund are saed in he parenheses. Significance levels are saed a 10%, 5% and 1% using *, **, and *** respecively. LISTED Variables Linear Non-Linear Linear Non-Linear Linear Non-Linear Linear Non-Linear Linear Non-Linear Consan ln(age) ln(size) Expense 30 Day Volailiy Syle Flow Linear Quadraic Raw Reurn Carhar - Adjused Fama French- Adjused CAPM - Adjused Benchmark - Adjused -0.0349*** -0.0339*** -0.0318*** -0.0324*** -0.0346*** -0.0349*** -0.0355*** -0.0353*** -0.0374*** -0.0393*** (0.0097) (0.0097) (0.0097) (0.0096) (0.0097) (0.0097) (0.0097) (0.0096) (0.0097) (0.0098) 0.0012 0.0013 0.0009 0.0011 0.0011 0.0012 0.0012 0.0013 0.0012 0.0013 (0.0008) (0.0008) (0.0008) (0.0008) (0.0008) (0.0008) (0.0008) (0.0008) (0.0008) (0.0008) 0.0008** 0.0007** 0.0008** 0.0007* 0.0008** 0.0007* 0.0008** 0.0007* 0.0009** 0.0009** (0.0004) (0.0004) (0.0004) (0.0004) (0.0004) (0.0004) (0.0004) (0.0004) (0.0004) (0.0004) 0.3113*** 0.2898*** 0.3688*** 0.5036*** 0.366*** 0.4913*** 0.3715*** 0.5116*** 0.2978*** 0.3033*** (0.1089) (0.1093) (0.1084) (0.1114) (0.1089) (0.1121) (0.1087) (0.1114) (0.1091) (0.1095) -0.1339-0.1228-0.1691* -0.0744-0.163* -0.0677-0.1676* -0.0509-0.1356-0.1367 (0.0949) (0.0947) (0.0945) (0.1007) (0.0947) (0.1023) (0.0946) (0.1041) (0.0958) (0.096) 0.0013* 0.0013* 0.0019*** 0.0019*** 0.0018*** 0.0017** 0.0017** 0.0017** 0.0012* 0.0012* (0.0007) (0.0007) (0.0007) (0.0007) (0.0007) (0.0007) (0.0007) (0.0007) (0.0007) (0.0007) 0.0069*** -0.0028 0.0154*** 0.0095** 0.0105*** 0.0079* 0.0115*** 0.0071 0.0009-0.0002 (0.002) (0.0046) (0.0019) (0.0041) (0.0019) (0.0046) (0.0019) (0.0045) (0.0018) (0.0047) - -0.0162** - -0.0308** - -0.0294*** - -0.0345*** 0.0112 0.0112 - (0.0077) - (0.0122) - (0.0085) - (0.0085) - (0.0071) - 0.0689** - -0.0018-0.0171-0.0289-0.0066
Table 8 Flow-Performance Funcional Form for Unlised Funds Table 8 conains he coefficiens of piecewise ime-fixed effec panel regression of he unlised sample o es he funcional form of he flow-performance relaionship. The dependen variable is oal fund flow, which is calculaed as he excess percenage growh of a fund ha would have occurred naurally. ln(age) is he logarihm of he fund s age, which is measured as he number of days since incepion and monh -1. ln(size) is he logarihm of he fund s size, expense is he ne expense raio of he fund for monh -1, 30 day volailiy is he sandard deviaion of daily reurns during he monh -1, and syle flow is he percenage flow ino fund i s Morningsar caegory. The funcional performance componen is represened by Linear, Quadraic, and Cubic variables; where Linear, Quadraic and Cubic represen he funcional form of he flowperformance relaionship of he fund s performance specified in he firs row of he able agains oher funds in he sample; specifically, he hree funcional forms are calculaed by he following Linear f, = min(rank f, 0.5), Quadraic f, = min(rank f, 0.5) 2, and Cubic f, = (Rank f, 0.5) 3. The sandard errors clusered by fund are saed in he parenheses. Significance levels are saed a 10%, 5% and 1% using *, **, and *** respecively. UNLISTED Variables Linear Non-Linear Linear Non-Linear Linear Non-Linear Linear Non-Linear Linear Non-Linear Consan ln(age) ln(size) Expense 30 Day Volailiy Syle Flow Linear Quadraic Cubic Raw Reurn Carhar - Adjused Fama French- Adjused CAPM - Adjused Benchmark - Adjused 0.0682*** 0.0685*** 0.0694*** 0.0676*** 0.0678*** 0.0648*** 0.0665*** 0.0634*** 0.0688*** 0.0698*** (0.0149) (0.0149) (0.0149) (0.015) (0.0149) (0.0149) (0.0149) (0.0149) (0.0151) (0.015) -0.0061*** -0.006*** -0.0061*** -0.006*** -0.0061*** -0.006*** -0.0062*** -0.006*** -0.0059*** -0.006*** (0.0011) (0.0011) (0.0011) (0.0011) (0.0011) (0.0011) (0.0011) (0.0011) (0.0011) (0.0011) -0.0011* -0.0011* -0.0012* -0.0012* -0.0011* -0.001-0.001-0.0009-0.0012* -0.0012* (0.0006) (0.0006) (0.0006) (0.0006) (0.0006) (0.0006) (0.0006) (0.0006) (0.0006) (0.0006) 0.165** 0.1642** 0.1783*** 0.1846*** 0.1926*** 0.1458** 0.1972*** 0.1593** 0.1862*** 0.1997*** (0.0643) (0.0643) (0.0647) (0.0648) (0.0644) (0.0657) (0.0643) (0.0655) (0.0652) (0.066) -0.407*** -0.3671*** -0.5018*** -0.5241*** -0.4609*** -0.5491*** -0.4679*** -0.5576*** -0.5264*** -0.5098*** (0.0916) (0.092) (0.0915) (0.092) (0.0904) (0.0955) (0.0902) (0.0974) (0.0922) (0.093) 0.2234*** 0.2193*** 0.2335*** 0.2294*** 0.2201*** 0.2108*** 0.2113*** 0.2015*** 0.205*** 0.2076*** (0.0655) (0.0654) (0.0651) (0.0652) (0.0651) (0.0652) (0.065) (0.0652) (0.0665) (0.0666) 0.0142*** 0.0138*** 0.0027-0.0038 0.0136*** 0.0078 0.0147*** 0.0112** 0.0029 0.0045 (0.0021) (0.0051) (0.0021) (0.0051) (0.002) (0.0048) (0.002) (0.0048) (0.002) (0.0052) - -0.0238*** - 0.0126-0.027*** - 0.0233*** - -0.0109 - (0.0081) - (0.0082) - (0.0088) - (0.009) - (0.0079) - 0.0063-0.0429-0.0374-0.0221 - -0.0093 - (0.0321) - (0.0321) - (0.0297) - (0.0301) - (0.0313) Observaions 6,610 6,610 6,610 6,610 6,610 6,610 6,610 6,610 6,471 6,471 R-Square 0.0196 0.0209 0.0128 0.0135 0.0192 0.0209 0.0203 0.0214 0.0123 0.0126 54
Table 9 Flow-Performance wih Ineracion Terms: Lised vs. Unlised Table 9 conains he coefficiens of piecewise ime-fixed effec panel regression of lised and unlised sample o es he robusness of our resuls. The dependen variable is oal fund flow, which is calculaed as he excess percenage growh of a fund ha would have occurred naurally. ln(age) is he logarihm of he fund s age, which is measured as he number of days since incepion and monh -1. ln(size) is he logarihm of he fund s size; expense is he ne expense raio of he fund for monh -1, 30 day volailiy is he sandard deviaion of daily reurns during he monh -1, and syle flow is he percenage flow ino fund i s Morningsar caegory. We added ineracion variables beween expense and he hree ranking quinile variables (Low f, Mid f, and High f,) as well as he an ineracion erm beween age and performance. Performance is defined as he reurn of he funds according o he firs row of he able and is calculaed on a monhly basis. The sandard errors clusered by fund are saed in he parenheses. Significance levels are saed a 10%, 5% and 1% using *, **, and *** respecively. Raw Reurn Carhar - Adjused Fama French - Adjused CAPM - Adjused Benchmark - Adjused Variables Lised Non-Lised Lised Non-Lised Lised Non-Lised Lised Non-Lised Lised Non-Lised Consan ln(age) ln(size) Expense 30 Day Volailiy Syle Flow Low*Expense Mid*Expense High*Expense Age*Perf -0.0360*** 0.0900*** -0.035*** 0.0892*** -0.0364*** 0.0887*** -0.037*** 0.0876*** 0.0145 0.0892*** (0.0097) (0.0197) (0.0096) (0.0199) (0.0097) (0.0199) (0.0097) (0.0199) (0.0776) (0.02) 0.0012-0.0079*** 0.001-0.0078*** 0.0012-0.0085*** 0.0012-0.0084*** -0.008-0.0076*** (0.0008) (0.0014) (0.0008) (0.0015) (0.0008) (0.0015) (0.0008) (0.0015) (0.0088) (0.0015) 0.0007** -0.0015* 0.0007** -0.0014 0.0008** -0.0014 0.0008** -0.0013 0.0023-0.0016* (0.0004) (0.0008) (0.0004) (0.0009) (0.0004) (0.0009) (0.0004) (0.0009) (0.0028) (0.0009) 0.0274-0.2727 0.2162 0.0511 0.4059 0.0762 0.2039 0.0584-0.1165-0.1019 (0.2061) (0.1872) (0.3225) (0.1851) (0.3158) (0.1985) (0.3057) (0.1963) (0.956) (0.1406) 0.1813* -0.2261* -0.0926-0.6201*** -0.1103-0.6632*** -0.0936-0.6463*** 0.4787-0.3691*** (0.1015) (0.1235) (0.0993) (0.1144) (0.1012) (0.1156) (0.1024) (0.1183) (0.3606) (0.1241) 0.0011* 0.2381*** 0.0018*** 0.2786*** 0.0017** 0.2766*** 0.0016** 0.2777*** 0.0002 0.2063** (0.0007) (0.0885) (0.0007) (0.0886) (0.0007) (0.0887) (0.0007) (0.0888) (0.0014) (0.0897) -0.0924 1.1498 0.7295 1.2588 0.2096 1.096 1.3583 0.8601-0.4161 1.4457* (1.0018) (0.9763) (1.3726) (1.0067) (1.3353) (0.8909) (1.2994) (0.88) (2.5742) (0.8005) 0.5028** 0.6162*** 0.4628* -0.2885 0.0069-0.351 0.0614-0.0682-0.8672-0.0338 (0.2142) (0.2285) (0.2782) (0.2192) (0.2773) (0.3664) (0.2642) (0.3644) (0.6693) (0.2118) -0.568-1.1128-2.7926** 0.3892-2.011 1.5724* -2.1793 0.7494 4.9742** -0.3956 (0.979) (0.8995) (1.2181) (0.9299) (1.2514) (0.8805) (1.3267) (0.8878) (2.535) (0.8115) 0.0118*** 0.0146*** 0.0153* 0.0214*** 0.0155** 0.0016* 0.0132* 0.0014* 0.0014 0.0112*** (0.0018) (0.0031) (0.0081) (0.0044) (0.0077) (0.0008) (0.0073) (0.0008) (0.0045) (0.002) Observaions 4,880 6,610 4,880 6,610 6,610 4,880 6,610 4,880 4,682 6,471 R-Square 0.0284 0.0222 0.0206 0.0159 0.0122 0.0154 0.0147 0.1541 0.0360 0.0154 55
Table 10 Flow-Performance wih Expense Dummy Ineracion Terms: Lised vs. Unlised Table 10 conains he coefficiens of piecewise ime-fixed effec panel regression of lised and unlised sample o es he funcional form of he robusness of our resuls. The dependen variable is oal fund flow, which is calculaed as he excess percenage growh of a fund ha would have occurred naurally. ln(age) is he logarihm of he fund s age, which is measured as he number of days since incepion and monh -1. ln(size) is he logarihm of he fund s size; expense is he ne expense raio of he fund for monh -1, 30 day volailiy is he sandard deviaion of daily reurns during he monh -1, and syle flow is he percenage flow ino fund i s Morningsar caegory. We added ineracion variables beween he hree ranking quinile variables (Low f, Mid f, and High f,) and an expense dummy. The expense dummy indicaes 1 if for a paricular observaion he expense is above he median of he sample, oherwise he dummy indicaes 0. The regression is conduced using differen performance measures. The sandard errors clusered by fund are saed in he parenheses. Significance levels are saed a 10%, 5% and 1% using *, **, and *** respecively. Raw Reurn Carhar - Adjused Fama French- Adjused CAPM - Adjused Benchmark - Adjused Variables Lised Unlised Lised Unlised Lised Unlised Lised Unlised Lised Unlised Consan ln(age) ln(size) Expense 30 Day Volailiy Syle Flow Low_EXP_D Mid_EXP_D High_EXP_D -0.0338*** 0.093*** -0.0337*** 0.0922*** -0.0351*** 0.0901*** -0.035*** 0.0896*** -0.0384*** 0.0904*** 0.0097 (0.02) (0.0096) (0.0201) (0.0097) 0.0201 (0.0097) 0.0201 (0.0097) 0.0202 0.001-0.0078*** 0.0009-0.0078*** 0.001-0.0077*** 0.001-0.0077*** 0.0012-0.0076*** (0.0008) (0.0015) (0.0008) (0.0015) (0.0008) 0.0015 (0.0008) 0.0015 (0.0008) 0.0015 0.001** -0.0016* 0.0011*** -0.0015* 0.001*** -0.0014* 0.0011*** -0.0014 0.0011*** -0.0015* (0.0004) (0.0009) (0.0004) (0.0009) (0.0004) 0.0009 (0.0004) 0.0009 (0.0004) 0.0009-0.2505 0.0641-0.3417** 0.1811-0.2015 0.1636-0.3042* 0.1706-0.0905 0.1966 (0.1709) (0.1296) (0.1623) (0.1277) (0.1641) 0.1226 (0.1642) 0.1217 (0.1705) 0.1223-0.0502-0.5594*** -0.111-0.6179*** -0.1132-0.6753*** -0.0981-0.6757*** -0.1181-0.627*** (0.0965) (0.1128) (0.0979) (0.1155) (0.0989) 0.1153 (0.1002) 0.117 (0.0955) 0.1131 0.0013** 0.2596*** 0.0016** 0.2632*** 0.0015** 0.2644*** 0.0014** 0.2662*** 0.0012* 0.2478*** (0.0007) (0.0891) (0.0007) (0.0884) (0.0007) 0.0888 (0.0007) 0.0886 (0.0007) 0.0891-0.0033-0.0304** 0.0301*** -0.0003 0.0305*** -0.0138 0.0384*** -0.0198 0.0221* -0.0144 (0.012) (0.0154) (0.0115) (0.0157) (0.0116) 0.0152 (0.0118) 0.0149 (0.0121) 0.0155 0.0222*** 0.0254*** 0.0131*** -0.0003 0.0056 0.0063 0.0058 0.0107* -0.0015 0.007 (0.0045) (0.0055) (0.0042) (0.0057) (0.0042) 0.0054 (0.0044) 0.0056 (0.0044) 0.0054 0.0219 0.0029-0.0226 0.0008-0.0081 0.043* -0.0139 0.0282 0.0597*** 0.0004 (0.0204) (0.0248) (0.0179) (0.0247) (0.0182) 0.0231 (0.018) 0.0235 (0.018) 0.0211 Observaions 4,880 6,610 4,880 6,610 4,880 6,610 4,880 6,610 4,770 6,610 R-Square 0.0191 0.0152 0.0146 0.0113 0.0095 0.0127 0.0114 0.0128 0.0102 0.0116 56
Table 11 Flow-Performance wih Size Dummy Ineracion Terms: Lised vs. Unlised Table 11 conains he coefficiens of piecewise ime-fixed effec panel regression of lised and unlised sample o es he funcional form of he robusness of our resuls. The dependen variable is oal fund flow, which is calculaed as he excess percenage growh of a fund ha would have occurred naurally. ln(age) is he logarihm of he fund s age, which is measured as he number of days since incepion and monh -1. ln(size) is he logarihm of he fund s size, expense is he ne expense raio of he fund for monh -1, 30 day volailiy is he sandard deviaion of daily reurns during he monh -1, and syle flow is he percenage flow ino fund i s Morningsar caegory. We added ineracion variables beween he hree ranking quinile variables (Low f, Mid f, and High f,) and a size dummy. The size dummy indicaes 1 if for a paricular observaion he size is above he median of he sample, oherwise he dummy indicaes 0. The regression is conduced using differen performance measures. The sandard errors clusered by fund are saed in he parenheses. Significance levels are saed a 10%, 5% and 1% using *, **, and *** respecively.. Raw Reurn Carhar - Adjused Fama French- Adjused CAPM - Adjused Benchmark - Adjused Variables Lised Unlised Lised Unlised Lised Unlised Lised Unlised Lised Unlised Consan ln(age) ln(size) Expense 30 Day Volailiy Syle Flow Low_SIZE_D Mid_SIZE_D High_SIZE_D -0.0321*** 0.1101*** -0.021 0.1104*** -0.0248* 0.1176*** -0.0218* 0.119*** -0.0368*** 0.0982*** 0.0124 (0.0282) (0.0129) (0.0281) (0.0129) (0.0283) (0.0129) (0.028) 0.0124 0.029 0.0012-0.0078*** 0.0009-0.0078*** 0.0011-0.0079*** 0.0011-0.0079*** 0.0013-0.0076*** (0.0008) (0.0015) (0.0008) (0.0015) (0.0008) (0.0015) (0.0008) (0.0015) (0.0008) 0.0015 0.0006-0.0025* 0.0001-0.0025* 0.0002-0.0029** 0.0001-0.0029** 0.0008-0.0019 (0.0005) (0.0013) (0.0006) (0.0013) (0.0006) (0.0013) (0.0006) (0.0013) (0.0005) 0.0013 0.3049*** 0.1668** 0.3178*** 0.1771** 0.3223*** 0.1769** 0.3135*** 0.1837** 0.3017*** 0.1744** (0.1093) (0.0809) (0.1082) (0.0811) (0.1089) (0.0811) (0.1088) (0.0811) (0.1092) 0.0816-0.1077-0.5823*** -0.1444-0.6276*** -0.1405-0.6569*** -0.1426-0.6368*** -0.1322-0.6365*** (0.0958) (0.113) (0.0957) (0.1128) (0.0967) (0.1137) (0.0965) (0.1143) (0.096) 0.1124 0.0012* 0.2559*** 0.0017** 0.2622*** 0.0016** 0.2661*** 0.0015** 0.267*** 0.0012* 0.2488*** (0.0007) (0.0884) (0.0007) (0.0884) (0.0007) (0.0884) (0.0007) (0.0883) (0.0007) 0.0888-0.0194* -0.0194-0.015 0.0229-0.0063-0.0009-0.0009-0.0051-0.0113-0.0043 (0.011) (0.0165) (0.0111) (0.0166) (0.0112) (0.016) (0.011) (0.0158) (0.0114) 0.0169 0.0138*** 0.0212*** 0.0178*** -0.0085 0.0104** 0.0095 0.0084** 0.0151** 0.0051 0.0051 (0.0041) (0.0067) (0.0044) (0.0067) (0.0045) (0.006) (0.0043) (0.0062) (0.0043) 0.0065 0.001-0.0118-0.0104 0.0528* -0.0018 0.051 0.0092 0.0167 0.0323* 0.016 (0.0199) (0.0294) (0.0214) (0.0295) (0.0212) (0.0327) (0.0215) (0.0327) (0.0173) 0.0287 Observaions 4,880 6,610 4,880 6,610 4,880 6,610 4,880 6,610 4,770 6,533 R-Square 0.0075 0.0138 0.0093 0.0121 0.0064 0.0136 0.0066 0.0135 0.0063 0.0117 57
Table 12 Flow-Performance wih Age Dummy Ineracion Terms: Lised vs. Unlised Table 12 conains he coefficiens of piecewise ime-fixed effec panel regression of lised and unlised sample o es he funcional form of he robusness of our resuls. The dependen variable is oal fund flow, which is calculaed as he excess percenage growh of a fund ha would have occurred naurally. ln(age) is he logarihm of he fund s age, which is measured as he number of days since incepion and monh -1. ln(size) is he logarihm of he fund s size, expense is he ne expense raio of he fund for monh -1, 30 day volailiy is he sandard deviaion of daily reurns during he monh -1, and syle flow is he percenage flow ino fund i s Morningsar caegory. We added ineracion variables beween he hree ranking quinile variables (Low f, Mid f, and High f,) and an age dummy. The age dummy indicaes 1 if for a paricular observaion he age is above he median of he sample, oherwise he dummy indicaes 0. The regression is conduced using differen performance measures. The sandard errors clusered by fund are saed in he parenheses. Significance levels are saed a 10%, 5% and 1% using *, **, and *** respecively. Raw Reurn Carhar - Adjused Fama French- Adjused CAPM - Adjused Benchmark - Adjused Variables Lised Unlised Lised Unlised Lised Unlised Lised Unlised Lised Unlised Consan ln(age) ln(size) Expense 30 Day Volailiy Syle Flow Low_AGE_D Mid_AGE_D High_AGE_D -0.0104 0.1104*** 0.002 0.1146*** -0.0068 0.1136*** -0.0013 0.1154*** -0.0163 0.1132*** 0.0122 (0.0236) (0.0123) (0.0236) (0.0124) 0.0236 (0.0122) 0.0234 0.0123 0.0238-0.0021* -0.0106*** -0.0038*** -0.0111*** -0.0027** -0.0111*** -0.0034*** -0.0114*** -0.0016-0.0109*** (0.0012) (0.0022) (0.0013) (0.0021) (0.0013) 0.0021 (0.0013) 0.0021 (0.0013) 0.0021 0.0007** -0.0014* 0.0007** -0.0014* 0.0008** -0.0014 0.0007** -0.0013 0.0009** -0.0015* (0.0004) (0.0009) (0.0004) (0.0009) (0.0004) 0.0009 (0.0004) 0.0009 (0.0004) 0.0009 0.3471*** 0.177** 0.449*** 0.1846** 0.4184*** 0.1884** 0.4361*** 0.1891** 0.3283*** 0.1873** (0.11) (0.0809) (0.1104) (0.081) (0.1105) 0.0814 (0.1105) 0.0812 (0.1099) 0.0814-0.1132-0.5671*** -0.1187-0.6133*** -0.1240-0.624*** -0.1143-0.6281*** -0.1399-0.6123*** (0.0959) (0.1144) (0.0967) (0.1134) (0.0973) 0.1155 (0.0972) 0.1166 (0.096) 0.1134 0.0012* 0.2549*** 0.0018*** 0.2599*** 0.0016** 0.2638*** 0.0016** 0.2645*** 0.0012* 0.2437*** (0.0007) (0.0887) (0.0007) (0.0883) (0.0007) 0.0885 (0.0007) 0.0886 (0.0007) 0.0891 0.0086-0.0018 0.0196* 0.027* 0.021* 0.008 0.0295*** 0.0137 0.0263** 0.0157 (0.0112) (0.0155) (0.0116) (0.0154) (0.0116) 0.0146 (0.0112) 0.0146 (0.0117) 0.0154 0.0116*** 0.0148** 0.0184*** -0.004 0.0103** 0.009 0.0099** 0.0065-0.0026 0.0052 (0.0042) (0.0061) (0.0044) (0.0061) (0.0045) 0.0056 (0.0043) 0.0057 (0.0045) 0.0061 0.0175-0.0081-0.0420** 0.0534* -0.0186 0.0309-0.0275 0.0399 0.0328* 0.0058 (0.0200) (0.0258) (0.0187) (0.0274) (0.0189) 0.0274 (0.0195) 0.0276 (0.0181) 0.025 Observaions 4,880 6,610 4,880 6,610 4,880 6,610 4,880 6,610 4,770 6,533 R-Square 0.0110 0.0130 0.0156 0.0126 0.0101 0.0132 0.0116 0.0133 0.0071 0.0122 58
Table 13 Flow-Performance Relaionship: Exchange Flow vs. Oher Flow Table 13 conains he coefficiens of piecewise ime-fixed effec panel regression of he lised and unlis es flow-performance relaionship of exchange flow and oher flow. The dependen variable is exchang (EX_FLOW), he amoun of flow ha originaes from he exchange raded channel, expressed as a perc Oher Flow (OFLOW) is he amoun of flow ha passes hrough all oher channel besides he exchang Independen variables include ln(age) is he logarihm of he fund s age, which is measured as he numb incepion and monh -1. ln(size) is he logarihm of he fund s size, expense is he ne expense raio of monh -1, 30 day volailiy is he sandard deviaion of daily reurns during he monh -1, and syle flow percenage flow ino fund i s Morningsar caegory. The performance componen is represened by Low variables; where Low, Mid and High represens he percenile rank of he fund s raw performance again same Morningsar caegory and region; specifically, he hree performance quiniles are calculaed by he = min(rank f,, 0.2), Mid f, = min(rank f, - Low f,, 0.2), and High f, = Rank f, - Low f, Mid f,.. To perform eses, we narrow he definiion of quiniles cu offs for ranking o 10% for Low and High quiniles; spe calculaed as he following: Low f, = min(rank f,, 0.1), Mid f, = min(rank f, - Low f,, 0.1), and High f, = R Mid f,.. Sandard errors clusered by fund are saed in he parenheses. Significance levels are saed a 1 using *, **, and *** respecively. Variables EX-FLOW OFLOW EX-FLOW OFLOW Consan ln(age) ln(size) Expense 30 Day Volailiy Syle Flow Low Mid High Low (20%) Mid (60%) High (20%) Low (10%) Mid (80%) High (10%) -0.0031* 0.0011-0.0031 0.0036 (0.0018) (0.0122) (0.0019) (0.0127) 0.0000-0.0001 0.0000-0.0001 (0.0001) (0.001) (0.0001) (0.001) 0.0001-0.0003 0.0001-0.0003 (0.0001) (0.0004) (0.0001) (0.0005) 0.0066-1.3375*** 0.0062-1.3388*** (0.02) (0.1256) (0.0199) (0.1254) 0.0711*** 2.2901*** 0.0709*** 2.2881*** (0.0213) (0.1365) (0.0213) (0.1365) -0.0002** -0.0011* -0.0002** -0.0011* (0.0001) (0.0006) (0.0001) (0.0006) 0.0012-0.0378** 0.0009-0.0929* (0.0032) (0.0191) (0.009) (0.0523) 0.001* 0.0103*** 0.0011** 0.0064** (0.0006) (0.0038) (0.0004) (0.0029) -0.0005-0.0224-0.0086-0.0561 (0.0028) (0.0166) (0.0079) (0.0431) Observaions 4,880 4,880 4,880 4,880 R-Square 0.0072 0.0969 0.0076 0.0966 59
Table 14 Flow-Performance Relaionship: Exchange Flow vs. Oher Flow Using Difference Performance Measures Table 14 conains he coefficiens of piecewise ime-fixed effec panel regression of he lised and unlised sample o es flow-performance relaionship of exchange flow and oher flow. The dependen variable is exchange flow (EX_FLOW) is he amoun of flow ha originaes from he exchange-raded channel, expressed as a percenage of TNA. Oher Flow (OFLOW) is he amoun of flow ha passes hrough all oher channel besides he exchange-raded channel. Independen variables include ln(age) is he logarihm of he fund s age which is measured as he number of days since incepion and monh -1. ln(size) is he logarihm of he fund s size, expense is he ne expense raio of he fund for monh -1, 30 day volailiy is he sandard deviaion of daily reurns during he monh -1, and syle flow is he percenage flow ino fund i s Morningsar caegory. The performance componen is represened by Low, Mid, and High variables; where Low, Mid and High represens he percenile rank of he fund s various performance measures agains peers in he same sample; specifically, he hree performance quiniles are calculaed by he following Low f, = min(rank f,, 0.2), Mid f, = min(rank f, - Low f,, 0.2), and High f, = Rank f, - Low f, Mid f,. The sandard errors clusered by fund are saed in he parenheses. Significance level are saed a 10%, 5% and 1% using *, **, and *** respecively. Variables EX-FLOW OFLOW EX-FLOW OFLOW EX-FLOW OFLOW EX-FLOW OFLOW EX-FLOW OFLOW Consan ln(age) ln(size) Expense 30 Day Volailiy Syle Flow Low Mid Raw Reurn Carhar - Adjused Fama French - Adjused CAPM - Adjused Benchmark - Adjused -0.0036* -0.0029-0.0034* 0.0034-0.0035* -0.001-0.0036* -0.0053-0.0028-0.0009 (0.0019) (0.0122) (0.0019) (0.0121) (0.0019) (0.0121) (0.0019) (0.012) (0.0018) 0.012 0.0000 0.0000 0.0000-0.0003 0.0000-0.0002 0.0000 0.0000-0.0001 0.0001 (0.0001) (0.001) (0.0001) (0.001) (0.0001) (0.001) (0.0001) (0.001) (0.0001) 0.001 0.0001-0.0003 0.0001-0.0003 0.0001-0.0003 0.0001-0.0003 0.0001-0.0004 (0.0001) (0.0004) (0.0001) (0.0004) (0.0001) (0.0004) (0.0001) (0.0004) (0.0001) 0.0005 0.0045-1.3396*** 0.0176-1.342*** 0.0122-1.3459*** 0.0123-1.3768*** 0.0058-1.3225*** (0.02) (0.1258) (0.021) (0.1305) (0.0212) (0.132) (0.021) (0.1327) (0.0203) 0.1271 0.0742*** 2.3068*** 0.0803*** 2.2565*** 0.0745*** 2.2298*** 0.0743*** 2.1797*** 0.0717*** 2.3052*** (0.0212) (0.1368) (0.0214) (0.1415) (0.0216) (0.1414) (0.0216) (0.1433) (0.0215) 0.1376-0.0002** -0.0012* -0.0002** -0.0007-0.0002** -0.0007-0.0002** -0.0009-0.0002** -0.001 (0.0001) (0.0006) (0.0001) (0.0006) (0.0001) (0.0006) (0.0001) (0.0006) (0.0001) 0.0007 0.0052-0.01 0.0028-0.0467*** 0.0046-0.0188 0.0051* 0.0096 0.0026-0.0146 (0.0033) (0.0186) (0.0028) (0.018) (0.0028) (0.0183) (0.0029) (0.0187) (0.003) 0.0178-0.0003 0.0034 0.0007 0.017*** -0.0001 0.0113*** -0.0002 0.0016 0.0008 0.0047 (0.0006) (0.0038) (0.0006) (0.0039) (0.0006) (0.0039) (0.0006) (0.0039) (0.0006) 0.0039
Table 15 Resul of Performance Predicabiliy Tes for Unlised Funds Table 15 conains he coefficiens of piecewise ime-fixed effec panel regression of he unlised sample o es he predicive abiliy of muual fund flows. The dependen variable conemporaneous performance measure for fund i as lised by he firs row of he able. Independen variables include ln(age) is he logarihm of he fund s age which is measured as he number of days since incepion and monh -1. ln(size) is he logarihm of he fund s size, expense is he ne expense raio of he fund for monh -1, and Toal flow is calculaed as he excess percenage growh of a fund ha would have occurred naurally. Reurn_LAG is he one monh lagged performance for fund i. The sandard errors clusered by fund are saed in he parenheses. Significance levels are saed a 10%, 5% and 1% using *, **, and *** respecively. Variables Lised Unlised Lised Unlised Lised Unlised Lised Unlised Lised Unlised Consan Reurn_LAG ln(age) ln(size) Expense Toal Flow Caegory-Adjused Carhar-Adjused Fama-French-Adjused CAPM-Adjused Benchmark-Adjused -0.3390 0.4250-0.7144 2.0913*** -0.8223 2.0772*** 0.6144 3.1829*** -1.6002-3.5644** (1.1326) (1.2909) (0.8211) (0.665) (0.8584) (0.6959) (0.9149) (0.7336) (1.29) (1.5014) -0.1277*** -0.1400*** -0.0621*** -0.0144-0.0629*** -0.0389*** -0.0848*** -0.0647*** -0.0140-0.2548*** (0.0149) (0.0129) (0.0232) (0.0166) (0.0234) (0.017) (0.0264) (0.0183) (0.0275) (0.019) 0.0959-0.0023 0.1062-0.0011 0.0897-0.0267-0.0415-0.0175 0.0458 0.1518 (0.0953) (0.0925) (0.0679) (0.0538) (0.0713) (0.056) (0.0748) (0.0579) (0.1076) (0.1154) -0.0048 0.0032-0.0029-0.0944*** 0.0115-0.0837*** -0.0054-0.1434*** 0.0529 0.1218** (0.0481) (0.0507) (0.0301) (0.0269) (0.0311) (0.0282) (0.033) (0.0295) (0.0583) (0.0604) -0.1029-0.1040* -0.2408*** -0.1406*** -0.2829*** -0.1597*** -0.2461*** -0.1653*** 0.1117 0.0492 (0.109) (0.0549) (0.0733) (0.0304) (0.0775) (0.0322) (0.0798) (0.0337) (0.1159) (0.0669) 0.0477*** -0.0172* 0.0121-0.0006 0.035** -0.0062 0.0439*** -0.0046 0.1285*** 0.0473*** (0.0123) (0.0099) (0.0144) (0.0049) (0.0152) (0.0052) (0.0169) (0.0054) (0.0154) (0.0112) Observaions 4,113 5,559 4,880 6,610 4,880 6,610 4,880 6,610 4,769 6,469 R-Square 0.0210 0.0207 0.0081 0.0047 0.0137 0.0062 0.0193 0.1036 0.0198 0.0625 61
Table 16 Resul of Performance Predicabiliy Tes for Lised Funds Table 16 conains he coefficiens of piecewise ime-fixed effec panel regression of he unlised sample o es he predicive abiliy of muual fund flows. The dependen variable: conemporaneous performance measure for fund i as lised by he firs row of he able. Independen variables include ln(age) is he logarihm of he fund s age, which is measured as he number of days since incepion and monh -1. ln(size) is he logarihm of he fund s size, expense is he ne expense raio of he fund for monh -1, and oal flow is calculaed as he excess percenage growh of a fund ha would have occurred naurally. Reurn_LAG is he one monh lagged performance for fund i. The sandard errors clusered by fund are saed in he parenheses. Significance levels are saed a 10%, 5% and 1% using *, **, and *** respecively. Variables Exchange-Fl. Oher Flow Exchange-Fl. Oher Flow Exchange-Fl. Oher Flow Exchange-Fl. Oher Flow Exchange-Fl. Oher Flow Consan Reurn_LAG ln(age) ln(size) Expense Exchange Flow Oher Flow Caegory-Adjused Carhar-Adjused Fama-French-Adjused CAPM-Adjused Benchmark-Adjused -0.1189-0.2038-0.7053-0.7039-0.7729-0.7702 0.7280 0.7317-1.2713-1.5980 (1.1327) (1.1329) (0.8216) (0.8216) (0.8593) (0.8593) (0.9176) (0.9177) (1.2974) (1.2975) -0.1259*** -0.1262*** -0.0735*** -0.0738*** -0.0926*** -0.0929*** -0.1203*** -0.1206*** -0.0448-0.0445 (0.0149) (0.0149) (0.0189) (0.0189) (0.019) (0.0191) (0.0208) (0.0209) (0.0276) (0.0276) 0.0998 0.0981 0.1088 0.1088 0.0945 0.0947-0.0395-0.0396 0.0543 0.0537 (0.0956) (0.0956) (0.0679) (0.0679) (0.0714) (0.0714) (0.0749) (0.0749) (0.1084) (0.1084) -0.0137-0.0126-0.0043-0.0044 0.0073 0.0072-0.0114-0.0116 0.0333 0.0330 (0.0482) (0.0482) (0.0302) (0.0302) (0.0312) (0.0312) (0.0332) (0.0332) (0.0587) (0.0588) -0.1170-0.1127-0.2469*** -0.2470*** -0.2996*** -0.2996*** -0.2650*** -0.2650*** 0.0787 0.0793 (0.1093) (0.1095) (0.0726) (0.0726) (0.0767) (0.0767) (0.0789) (0.0789) (0.117) (0.117) 0.0083-0.0004*** - 0.0004*** - 0.0005*** - 0.0009** - 0.0065-0.0001 - (0.0002) - (0.0002) - (0.0003) - - 0.0014 - -0.0004*** - -0.0004** - -0.0005*** - -0.0005* - (0.0079) - 0.0001 - (0.0002) - (0.0002) - (0.0003) Observaions 4,113 4,113 4,880 4,880 4,880 4,880 4,880 4,880 4,769 4,769 R-Square 0.0175 0.0169 0.0080 0.0080 0.0119 0.0118 0.0167 0.0166 0.0024 0.0022 62
Table 17 Resul of Facor Model Regressions Table 17 shows he Fama-Macbeh coefficiens and sandard errors for Carhar four facor, Fama-French hree facor and he CAPM one facor rolling regressions compleed using a hiry-six monh window. The lengh of he window was chosen on he basis ha i was consisen wih curren lieraure; see Ferreira e al. (2012) and Carhar (1997). 4 Facor Model 3 Facor Model 1 Facor Model Inercep RM SMB HML MOM ALPHA Inercep RM SMB HML ALPHA Inercep RM ALPHA Mean -0.0171 0.6119 0.1321-0.1152 0.0421-0.00389 0.0540 0.6125 0.2018-0.0668 0.1317% 0.1148 0.6658 0.7261% S.D 0.6017 0.4897 0.2834 0.4361 0.2784 2.921 0.6009 0.4860 0.3095 0.4394 3.0873 0.6013 0.4995 3.2664 PVALUE 0.4185 0.1090 0.5035 0.4085 0.3718 0.4223 0.1138 0.4417 0.4003 0.4190 0.0927 T-STAT 0.1491 3.8691 0.4301-0.3124-0.1583 0.2813 3.9688 0.6433-0.1708 0.4068 5.0788 RSQ 50.77% 47.49% 42.49% OBS 22,778 22,819 22,901 4 Facor Model 3 Facor Model 1 Facor Model Inercep RM SMB HML MOM ALPHA Inercep RM SMB HML ALPHA Inercep RM ALPHA Mean 0.0181 0.3938 0.0922-0.0296-0.0073-0.0003 0.0748 0.4005 0.1311 0.0118 0.2184% 0.1185 0.4365 0.4224% S.D 0.4783 0.406 0.2161 0.309 0.195 2.1629 0.4775 0.4044 0.2371 0.2995 2.2879 0.4762 0.4214 2.402612 PVALUE 0.4018 0.1296 0.5252 0.4041 0.3526 0.3974 0.1317 0.4750 0.4058 0.3870 0.1031 T-STAT 0.1628 3.2557 0.3969-0.1028-0.5259 0.2873 3.3993 0.5367 0.1154 0.4084 4.4032 RSQ 46.83% 43.31% 38.42% OBS 22,935 23,124 23,501 63
Table 18 Descripive Saisics for Lised Funds Table 18 shows he descripive and flow saisic for lised funds. Age is calculaed as he number of days beween incepion and he observaion dae; size is calculaed as he capializaion of he fund, expense is he ne expense raio for he fund, syle flow is he ne average percenage flow ino he designaed Morningsar caegory, exchange, oal and oher flow are ne percenage flows in and ou of funds during he monh. Flow definiions and calculaion mehodology can be found in he mehodology secion. Reurn is he raw monhly reurn of he fund, facor models are calculaed using a hiry-six rolling regression model, caegory reurn is calculaed by raw reurn minus caegory reurn, benchmark is calculaed as raw reurn minus benchmark reurn. All flow daa showed significan sandard deviaion and spread, hence we believe winsorizing flow variables may beer serve our invesigaion. Variable No. Mean Median Skew Kurosis Sd.Dev. Max Min Age(Days) 32,855 5,829 4,426 1.2873 0.9620 4,590 22,568 0.0000 Size ( m) 14,149 790.49 255.1 3.9115 17.1235 1,516 12,787 7.9981 Expense (%) 23,466 1.3931 1.4300 3.8251 51.627 0.6461 12.620 0.0900 6 Monh Reurn Volailiy (%) 31,871 3.9512 3.5118 1.0709 1.8081 2.9746 25.8344 0.0082 Syle Flow (%) 10,811 1.8812-0.0117 25.3421 746.211 32.3401 1,113.56-2.347 Toal Flow (%) 21,383 0.1108-0.0059 102.1996 10,518 10.7853 1,123.92-0.999 Exchange Flow ( m) 21,382-40.462-0.3738-110.7506 12,889 4,198 38,389-525,473 Oher Flow (%) 20,747-0.2026-0.4063 62.2046 11,218 611.00 72,903-45,138 Reurn (%) 32,685 0.4870 0.4100-0.5881 4.9019 4.9466 33.490-40.060 Fama-French 4 Facor Adj-Reurn (%) 22,442-0.0355 0.0229-0.1230 4.7934 2.9264 28.0897-23.4492 Fama-French 3 Facor Adj-Reurn (%) 22,483 0.0062 0.0355-0.0929 4.4351 3.0919 28.4459-24.0870 CAPM 1 Facor Adjused Reurn (%) 22,565 0.0538 0.0358 0.2474 4.7533 3.2694 27.8942-21.8003 Caegory Adjused Reurn (%) 26,415-0.1953-0.2872 0.3588 3.1631 3.7478 54.3935-20.5866 Benchmark Adjus Reurn (%) 32,252-0.0844-0.0637 0.5699 9.4489 3.8402 37.8568-33.0600 64
Table 19 Descripive Saisics for Unlised Funds Table 19 shows he descripive and flow saisic for lised funds. Age is calculaed as he number of days beween incepion and he observaion dae; size is calculaed as he capializaion of he fund, expense is he ne expense raio for he fund, syle flow is he ne average percenage flow ino he designaed Morningsar caegory, exchange, oal and oher flow are ne percenage flows in and ou of funds during he monh. Flow definiions and calculaion mehodology can be found in he mehodology secion. Reurn is he raw monhly reurn of he fund, facor models are calculaed using a hiry-six rolling regression model, caegory reurn is calculaed by raw reurn minus caegory reurn, benchmark is calculaed as raw reurn minus benchmark reurn. Variable No. Mean Median Skew Kurosis Sd.Dev. Max Min Age(Days) 35,948 2,980 2,301 1.6986 3.5212 2,637 15,733 0.0000 Size ( m) 33,786 760.51 302 5.1864 34.9762 1,492 19,124 0.0032 Expense (%) 24,042 1.4313 1.2100 1.9214 8.784 1.0387 11.640 0.0300 6 Monh Reurn Volailiy (%) 36,742 2.8622 2.2407 1.4222 4.3863 2.3988 31.0553 0.0000 Syle Flow (%) 10,324 14.7746-0.0229 14.7035 241.255 157.2950 3,018.49-3.000 Toal Flow (%) 27,003-1.4770 0.0000-157.8964 25,479.30 271.880 509.49-44025.501 Reurn (%) 36,058 0.4137 0.3536-0.5012 8.3150 3.7816 66.293-33.830 Fama-French 4 Facor Adj-Reurn (%) 22,216-0.0009 0.0258-0.3338 3.9641 2.1805 15.1321-16.0487 Fama-French 3 Facor Adj-Reurn (%) 22,396 0.0207 0.0272-0.2877 3.9611 2.3071 15.937-15.449 CAPM 1 Facor Adjused Reurn (%) 22,755 0.0371 0.0332 0.0253 4.1603 2.4230 19.0440-15.6758 Caegory Adjused Reurn (%) 28,224-0.1151-0.2542 0.6058 6.8319 3.6953 61.0931-20.6902 Benchmark Adjus Reurn (%) 12,333 0.0913 0.0291-0.2447 5.2293 0.0445 34.6200-29.4346 65
Table 20 Correlaion Co-efficien for Lised Funds Table 20 consiss of correlaion co-efficien used for flow-performance and performance predicabiliy sudies for lised funds. A lis of definiions for each variable can be found in appendix par B. The full marix is no prined here due o is size. I can be made available a he reader s reques. The omied variables consis of he low, mid, and high rankings for hree facor, one facor, and benchmark reurns. The magniude and direcion of he correlaion co-efficien beween he omied variables and descripive saisics are similar o he low, mid, and high rankings for raw, Morningsar and four facor-adjused reurns. LN_AGE LN_SIZE EXPENSE SFLOW TOTAL EXFLOW OFLOW MS_VOL LOW MID HIGH MC_LOW MC_MID MC_HIGH 4_LOW 4_MID 4_HIGH LN_AGE 1.0000 LN_SIZE 0.1971 EXPENSE -0.0403-0.2402 W_SFLOW 0.1124 0.0610 0.0736 W_TOTAL 0.0300 0.0298 0.0309 0.0376 W_EXFLOW -0.0025 0.0038 0.0234-0.0269-0.0072 W_OFLOW 0.0020-0.0352-0.0334-0.0400-0.1490-0.1481 MS_VOL 0.0009-0.1884 0.3579-0.0265-0.0128 0.0690 0.2739 LOW -0.0046 0.0119 0.0431 0.0256 0.0712 0.0234-0.0148-0.0100 MID 0.0300 0.0210 0.0183 0.0287 0.0493-0.0017 0.0027 0.0255 0.5394 HIGH 0.0420 0.0271 0.0173-0.0440 0.0306-0.0146-0.0048 0.0485 0.1719 0.5419 MC_LOW 0.0030 0.0265-0.0076-0.0074 0.0649 0.0235-0.0213-0.0304 0.3453 0.3428 0.1438 MC_MID -0.0041-0.0021 0.0050-0.0243 0.0666 0.0399 0.0339 0.0401 0.3015 0.4659 0.2805 0.5141 MC_HIGH -0.0253-0.0336 0.0351 0.0002 0.0329 0.0213 0.0179 0.0720 0.1482 0.3286 0.3128 0.1640 0.5450 4_LOW 0.0065 0.0819-0.2086-0.0398 0.1057 0.0137-0.0336-0.1837 0.4179 0.3732 0.1738 0.2881 0.2811 0.1369 4_MID 0.0508 0.0133-0.0541-0.1154 0.1123 0.0242 0.0697 0.0380 0.3018 0.4841 0.3796 0.3038 0.4322 0.2804 0.5407 4_HIGH 0.0203-0.0915 0.2019-0.0354 0.0431 0.0082 0.0735 0.2683 0.1672 0.4035 0.4991 0.1355 0.2853 0.2748 0.1761 0.5428 1.0000 66
Table 21 Correlaion Co-efficien for Unlised Funds Table 21 consiss of correlaion co-efficien used for flow-performance and performance predicabiliy sudies for unlised funds. A lis of definiions for each variable can be found in Appendix Par B. The full marix is no prined here due o is size. I can be made available a he reader s reques. The omied variables consis of he low, mid, and high rankings for hree facor, one facor, and benchmark reurns. The magniude and direcion of he correlaion co-efficien beween he omied variables and descripive saisics are similar o he low, mid, and high rankings for raw, Morningsar and four facor-adjused reurns. LN_AGE LN_SIZE EXPENSE SFLOW TOTAL MS_VOL LOW MID HIGH MC_LOW MC_MID MC_HIGH 4_LOW 4_MID 4_HIGH LN_AGE 1.0000 LN_SIZE 0.0361 EXPENSE -0.0996-0.1054 SFLOW -0.0665 0.0278-0.0039 TOTAL -0.0748-0.0272 0.0155 0.0465 VOL -0.0243-0.0555 0.3356-0.0797-0.0523 LOW -0.0047-0.0076-0.0286 0.0291 0.0724-0.1650 MID 0.0020-0.0083-0.0203 0.0414 0.0810-0.1369 0.5298 HIGH -0.0045 0.0082 0.0308-0.0161 0.0359 0.0007 0.1721 0.5448 MC_LOW 0.0509 0.0046-0.0856-0.0117 0.0583-0.0594 0.0234 0.0262-0.0118 MC_MID 0.0191 0.0310-0.0732-0.0093 0.0477-0.0316 0.0126 0.0208-0.0014 0.5094 MC_HIGH 0.0013 0.0017 0.0148-0.0242 0.0382 0.0076-0.0117 0.0238 0.0307 0.1542 0.5191 4_LOW 0.0349-0.0248 0.0109-0.0311 0.0247-0.0302 0.0030-0.0097-0.0181 0.0236 0.0199-0.0050 4_MID -0.0040-0.0348 0.0092-0.0190 0.0136 0.0512-0.0193-0.0097-0.0080 0.0115 0.0336 0.0145 0.5423 4_HIGH -0.0316-0.0277-0.0013-0.0032 0.0093 0.1168-0.0172 0.0144 0.0208 0.0159 0.0204 0.0032 0.1765 0.5424 1.0000 67