The Impact of Portfolio Disclosure on Hedge Fund. Performance, Fees, and Flows. Zhen Shi

Transcription

1 The Impac of Porfolio Disclosure on Hedge Fund Performance, Fees, and Flows by Zhen Shi A Disseraion Presened in Parial Fulfillmen of he Requiremens for he Degree Docor of Philosophy Approved April 2011 by he Graduae Supervisory Commiee: Michael Herzel, Co-Chair George Aragon, Co-Chair Jeffrey Coles ARIZONA STATE UNIVERSITY May 2011

2 ABSTRACT This sudy invesigaes he impac of porfolio disclosure on hedge fund performance. Using a regression disconinuiy design, I invesigae he effec of he disclosure requiremens ha ake effec when an invesmen company's asses exceed $100 million; when ha occurs, a fund is required by he SEC o submi form 13F disclosing is porfolio holdings. Consisen wih he argumen ha porfolio disclosure reveals "rade secres" and also raises fron running coss hus harms he funds ha disclose, I find ha here is a drop in fund performance (abou 4% annually) afer a fund begins filing form 13F, as well as an increase in reurn correlaions wih oher hedge funds in he same invesmen syle. The drop in performance canno be explained by a change in he asses under managemen or a mean reversion in reurns. Consisen wih he idea ha funds wih illiquid holdings end o employ sequenial rading sraegies, which increase he likelihood of being aken advanage of by free riders and fron runners, he drop in performance is more dramaic for funds ha have more illiquid holdings. In addiion, I find ha he incenive fees paid o fund managers are 1% higher when porfolio disclosure is required, which suppors he hypohesis ha invesors' monioring of porfolio holdings disciplines adverse risk-aking by fund managers and allows for higher convexiy in he opimal compensaion srucure. Finally, here is a drop in flows ino funds ha file 13F, which suggess ha hedge fund invesors negaively value 13F disclosure. Overall, his sudy suggess ha he cos of porfolio disclosure is economically large. I conribues o he policy debae over wha consiues opimal disclosure. i

3 ACKNOWLEGEMENT I am graeful o my advisors, George Aragon, Michael Herzel, and Jeffery Coles, for heir guidance, encouragemen, paience, and suppor. I hank my family for always being here for me and for supporing me hroughou my Ph.D. sudy. ii

4 TABLE OF CONTENTS Page LIST OF TABLES... vi CHAPTER 1 INTRODUCTION DATA... 7 Insiuional Invesmen Managers and Form 13F... 7 Sample Selecion... 7 Daes on Which Funds Sar or Sop Filing Form 13F EMPIRICAL RESULTS The Impac of 13F Disclosure on Hedge Fund Performance Hedge Fund Performance Before and During 13F Porfolio Disclosure Regression Disconinuiy (RD) Design Disconinuous Change in Performance When Funds Sar o File 13F Illiquidiy and he Impac of Porfolio Disclosure on Performance The Impac of 13F Disclosure on Correlaions of Hedge Fund Reurns The Impac of 13F Disclosure on Hedge Fund Compensaion Schemes Summary Saisics on Hedge Fund Compensaion Schemes. 35 iii

5 CHAPTER...Page The Impac of 13F Disclosure on Compensaion Schemes The Impac of 13F Disclosure on Hedge Fund Flows CONCLUSIONS REFERENCES APPENDIX A Probi Analysis on he Likelihood ha Firm Files 13F B The Impac of Porfolio Disclosure on Hedge Fund Risk iv

6 LIST OF TABLES Table Page 1. The Number of Companies Added or Dropped from he Lis of 13F Filing Companies The Lengh of 13F Filing Periods Hedge Fund Performance Before and During 13F Porfolio Disclosure Calendar Time Porfolios Regression Disconinuiy Design Change in Performance When Funds Sar o File 13F Illiquidiy and he Impac of Porfolio Disclosure on Hedge Fund Performance The Impac of 13F Disclosure on he Correlaions of Hedge Fund Reurns Summary Saisics on Hedge Fund Compensaion The Impac of 13F Disclosure on Hedge Fund Compensaion Schemes The Impac of 13F Disclosure on Fund Flows v

7 CHAPTER 1 INTRODUCTION Deermining he exen o which invesmen porfolios should be publicly disclosed is a basic challenge facing hedge fund indusry paricipans and regulaors. Porfolio disclosure is beneficial o he exen ha i allows invesors o make informed invesmen allocaion decisions and reduces poenial agency coss ha can arise when managerial acions are more opaque. Porfolio disclosure, however, is cosly if i reveals proprieary informaion and faciliaes free-riding aciviies by ohers on a fund's profiable invesmens and rading sraegies. 1 In his sudy, I invesigae he effec of he porfolio disclosure requiremens ha ake effec when an invesmen company's asses exceed $100 million; when ha occurs, hedge fund and oher insiuional managers are required by he SEC o file form 13F reporing heir quarerly holdings wihin 45 days afer he end of each quarer. This disconinuous change in disclosure regimes around he $100 million hreshold allows for he use of a regression disconinuiy approach and he idenificaion of a causal effec of porfolio disclosure on hedge fund performance ha is purged of poenial endogeneiy problems. The idenifying assumpion is ha he funcion ha relaes fund size o performance does no have precisely he same jumps as he funcion ha relaes fund size o disclosure. This procedure is valid even if unobserved facors ha affec performance (such as a fund manager's 1 Poerba, Shackelford, and Shoven (2004) demonsrae ha hypoheical copyca" funds creaed by mimicking he porfolio holdings of acively managed muual funds earn afer expense reurns ha are indisinguishable from he copied funds. 1

8 skills) are funcionally relaed o fund size. Using a complee sample of 4,024 hedge fund managers ha repor o TASS over he period of February 1977 o February 2010, among which 414 have led Form 13F a leas once, I find ha fund performance is lower in he disclosure periods han in he non-disclosure periods. The resuls are robus o five performance measures, including raw reurns, marke model alpha, Fama-French hree-facor alpha, Carhar four-facor alpha, and Fung-Hsieh seven-facor alpha. The cos of performance disclosure is economically large. For example, when measured by he Fama-French hree-facor alpha, he performance is 4% lower (annually) during he disclosure periods. I also find ha he drop in performance does no occur slowly over ime; insead, i occurs in he form of a sudden drop in he firs year afer a fund files is firs 13F disclosure. This finding of a disconinuous drop in performance along he ime dimension lends srong suppor o he argumen ha he decreased performance is due o porfolio disclosure. Using a regression disconinuiy design where samples are narrowed o a small neighborhood around he $100 million hreshold, I find ha here is a drop in fund performance ha occurs in he form of a disconinuous jump and ha i canno be explained by coninuous changes in he asses under managemen when he regulaory regime swiches from non-disclosure o disclosure. In addiion, I find ha here is no such disconinuous drop in performance for funds ha also 2

9 crossed he $100 million hreshold bu were no required o file 13F. 2 Furhermore, here is no drop in fund performance when funds crossed oher hresholds (e.g., $80 million or $120 million). These resuls confirm ha he drop in performance following disclosure is no due o change in size or mean reversion in reurns. In suppor of he hypohesis ha he decreased performance in he periods of porfolio disclosure is due o free-riding aciviies by oher fund managers, I find ha he reurn correlaions beween disclosing funds and oher hedge funds ha are in he same invesmen syle are greaer in he disclosure periods han in he non-disclosure periods. This finding suppors he idea ha oher funds ake posiions more similar o disclosing funds afer disclosing funds disclose heir porfolio holdings or ha incenives o pursue novel sraegies diminish following disclosure. I also invesigae he exen o which he liquidiy of porfolio holdings affecs he cos of disclosure. In general, rades in illiquid securiies resul in larger price impacs han rades in liquid securiies. In order o reduce he ransacion cos due o price impac, fund managers end o employ sequenial rading sraegies o accumulae or dispose of an illiquid posiion. However, he longer i akes o accumulae or dispose of a posiion, he higher he likelihood and greaer he cos of being aken advanage of by fronrunners and free riders. Consisen wih hese observaions, I find ha he drop in performance is more 2 The fac ha hese large funds are no required o file 13F is because hey hold non-13(f) securiies, shor posiions, or securiies ha are less han 10,000 shares or wih marke value less han $200,000. 3

10 dramaic for funds wih illiquid holdings. Disclosure is inended o improve monioring and reduce agency problems. One of he main agency problems facing hedge fund invesors is ha managers may ake on excessive risk, especially given he prevailing opion-like bonus" incenive fee. 3 Under bonus" incenive fees, hedge fund managers receive a fixed percenage of fund profi bu are no penalized when hey incur losses. Thus, hedge fund managers do no suffer any downside risk. Unlike he ime series of pas fund reurns, which provide a very limied view of fund risk, 4 porfolio disclosure allows invesors o observe he holdings and assess he risk ha hey are exposed o. Their monioring of porfolio holdings may discipline risk aking by fund managers and reduce he convexiy of he incenive fees. Consisen wih his argumen ha porfolio disclosure allows higher convexiy in he opimal compensaion srucure, I find ha incenive fees are 1% higher in he presence of porfolio disclosure, afer conrolling for oher facors such as asses under managemen and age of he fund families. Finally, wheher hedge fund invesors value 13F porfolio disclosure is sill an open quesion. While invesors may prefer more disclosure for he 3 Sarks (1987) and Grinbla and Timan (1989) show in heoreical models ha managers have incenives o choose greaer risks han he desired risk level by he cliens under opion like incenive fees. Coles, Daniel, and Naveen (2006) show empirically ha higher sensiiviy of CEO wealh o sock volailiy (vega) implemens riskier policy choices. Golec and Sarks (2004) find ha an exogenous change in incenive fees reduces muual fund managers' risk aking. 4 See Sulz (2007) for a discussion on earhquake" risks ha can' be deeced from pas performance. 5 See Healy and Palepu (2001) for a review of empirical disclosure lieraure. 4

11 increased ransparency i affords, hey may prefer less disclosure if i leads o lower fund performance. The value ha invesors aach o he 13F disclosure can be measured by he fund flows in and ou of he funds. I find ha flows are lower in he periods of disclosure han in non-disclosure periods. The es for his conrols for he change in flows ha migh be expeced in response o oher facors, such as changes in pas performance (Berk and Green, 2004). My finding suggess ha hedge fund invesors place a negaive value on 13F porfolio disclosure. This sudy is relaed o a broader lieraure on financial informaion disclosure. 5 There are several advanages of using hedge funds as a laboraory o examine issues relaed o disclosure. Firs, he proprieary cos of disclosure is plausibly more imporan for hedge fund managers. Hedge funds are relaively unfeered in heir abiliy o use leverage, derivaives, and shor sales across several asse classes. This srucure migh arac alened managers wih sophisicaed rading sraegies. Second, hedge funds ofen uilize lockup provisions and hold illiquid asses, pracices ha sugges hey are also more likely o use dynamic rading sraegies. Disclosure especially undermines he profiabiliy of hese sraegies. Third, he exen of disclosure, firm performance, and he value invesors aach o he disclosure policy can be direcly and easily measured in he conex of porfolio disclosure. In conras, he difficuly of measuring he exen of disclosure has consrained research in he area of financial informaion disclosure (Healy and Palepu, 2001). This sudy provides direc evidence ha porfolio disclosure harms hedge 5

12 fund performance and suggess ha he cos of disclosure is economically large. This finding is suppored by Aragon, Herzel, and Shi (2009) and by Agarwal, Jiang, Tang, and Yang's (2009), who demonsrae ha hedge fund managers reques confidenial reamen o delay 13F disclosure of heir profiable ideas. This sudy is also in line wih he finding of Ge and Zheng (2006) ha pas winner" muual funds ha disclose less frequenly ouperform hose ha disclose more frequenly. This sudy is also he firs o analyze he ineracions beween porfolio disclosure and compensaion srucure in he invesmen fund indusry. A fund manager's adverse risk-aking incenive is similar o he risk-shifing incenive of an equiy holder o expropriae wealh from exising bondholders. My finding ha incenive fees are higher in he presence of porfolio disclosure is similar o John, Mehran, and Qian's (2008) finding ha he pay-for-performance sensiiviy of CEO compensaion increases wih he inensiy of ouside monioring of he firm's risk choice, hough heir focus is no on he convexiy of he compensaion. The remainder of he paper is organized as follows. Secion II describes he daa. Secion III discusses he mehodology and empirical resuls. Secion IV concludes. 6

13 CHAPTER 2 DATA Insiuional Invesmen Managers and Form 13F Since 1978, all insiuional invesmen managers (including hedge fund managers) who exercise invesmen discreion over $100 million or more have been required by Secion 13(f) of he Exchange Ac o make quarerly disclosures of porfolio holdings o he SEC on form 13F. Form 13F mus be filed wih he SEC no laer han 45 days afer he end of each calendar quarer. The ypes of securiies ha are repored on form 13F include exchange-raded and NASDAQquoed socks, equiy opions and warrans, converible bonds, and shares of closed-end invesmen companies. All long posiions in such securiies wih more han 10,000 shares or wih a marke value exceeding $200,000 are required o be repored. Informaion repored on form 13F includes he issuers of he securiies, he securiy ype, he CUSIP number, he number of shares, and he marke value of each securiy owned. Managers are allowed o repor aggregaed holdings for differen funds managed by he same managemen company. Sample Selecion The Lipper/TASS hedge fund daabase provides monhly fund reurns and asses under managemen, a snapsho of fund characerisics, and he managemen company/invesmen advisor volunarily repored by hedge funds. The TASS hedge fund daabase repors daa beginning in February 1977, and he mos recen download covers daa o February A ha ime here were 13,845 funds, including 5,861 live funds and 7,984 dead funds. A oal of 4,024 managemen 7

14 companies/invesmen advisors are lised in he TASS daabase and each managemen company can manage muliple funds. The Thomson-Reuers Insiuional (13f) Holdings daase provides quarerly holdings by insiuional invesors ha are obligaed o file form 13F wih he SEC. The Thomson-Reuers daase sars from he firs quarer of 1980, and he mos recen downloads cover holdings unil he las quarer of In order o idenify invesmen companies ha manage hedge funds, I firs compile a lis of hedge fund company names using he Company file in he TASS Hedge Fund daases downloaded in February This yields a oal of 4,024 invesmen companies ha manage hedge funds. I hen hand mached hese hedge fund company names wih company names in he Thomson-Reuers Insiuional (13f) Holdings daase. There are a oal of 414 invesmen companies mached. Daes on Which Funds Sar or Sop Filing Form 13F The firs quarer ha a company has fling records in he Thomson (13f) daase is idenified as he quarer ha an invesmen company sars o file form 13F. Similarly, he las quarer ha a company has filing records in Thomson (13f) is idenified as he quarer ha an invesmen company sops filing form 13F. Table 1 liss he number of invesmen companies ha began filing form 13F (and were added o he daabase of 13F filing companies) each year from 1980 o 2009 in column 1 and he number of invesmen companies ha sopped filing form 13F (and were dropped from he 13F lis) in column 2. As shown in column 1, he number of invesmen companies added o he 13F lis increased over he firs half of he sample period and peaked in year I sayed roughly sable in he second half of he sample period and dropped in year Table 2 repors he 8

15 disribuion of he lengh of 13F filing periods. The majoriy of he invesmen companies have a filing period of beween 2 years and 10 years. There are hree invesmen companies ha have a filing period of over 20 years. 9

16 Table 1. The Number of Companies Added or Dropped from he Lis of 13F Filing Companies This able repors he number of invesmen companies ha began filing form 13F (added o he lis of 13F Filing companies) each year from 1980 o 2009 in column 1 and he number of invesmen companies ha ceased filing form 13F (and were dropped from he lis of 13F companies) in column 2. The firs quarer ha a company has filing records in he Thomson (13f) daase is idenified as he quarer ha an invesmen company sars o file form 13F, excluding he beginning of he Thomson (13f) daase, which is he firs quarer of year Similarly, he las quarer ha a company has filing records in Thomson (13f) is idenified as he quarer ha an invesmen company sops filing form 13F, excluding he las dae of he Thomson (13f) daa download, which is he las quarer of year Year # of Companies Tha Begin Filing 13F # of Companies Tha Cease Filing 13F Toal

17 Table 2. The Lengh of 13F Filing Periods This able repors he disribuion of he lengh of he 13F filing periods. Lengh of 13F Filing Period # of Invesmen Companies 1 quarer 24 2 quarers o 1 Year 52 2 o 5 Years o 10 Years o 15 Years o 20 Years 1 >= 20 Years 3 Toal

18 CHAPTER 3 EMPIRICAL RESULTS The Impac of 13F Disclosure on Hedge Fund Performance Hedge Fund Performance Before and During 13F Porfolio Disclosure In his secion, I use univariae ess o invesigae wheher porfolio disclosure harms hedge fund performance. Only hose fund families ha have reurn daa in TASS before and during 13F porfolio disclosure are included in he analysis. The saisical significance of he difference in performance beween disclosure and non-disclosure periods is obained using a paired -es. As shown in Table 3, hedge fund performance is worse in disclosure periods han in nondisclosure periods. The resuls are robus o five performance measures including raw reurns, marke model alpha, Fama-French hree-facor alpha, Carhar fourfacor alpha, and Fung-Hsieh seven-facor alpha. The differences in performance are boh saisically and economically significan. For example, he Fung-Hsieh seven-facor model alpha is 0.399% lower per monh (4.788% annually) during he 13F disclosure period. Risk-Adjused Performance of Calendar Time Porfolio I invesigae wheher a calendar ime porfolio which long disclosed funds and shor non-disclosed funds earns abnormal risk-adjused reurns. In each monh, a fund family is classified as disclosed if i files 13F in monh -1, oherwise i is classified as non-disclosed. Only he fund families ha file 13F a leas once during he sample period are included in he analysis. Table 4 repors raw reurns and risk adjused performance including marke model alpha, 12

19 Fama-French hree-facor alpha, Carhar four-facor alpha, and Fung-Hsieh sevenfacor alpha of he calendar ime porfolio. As shown in he able, raw reurns and alphas obained from he four risk models are all negaive and saisically significan. For example, he alpha based on Fung-Hsieh seven-facor model is % monhly (3.396% annually). These resuls sugges ha disclosed funds underperformance non-disclosed funds by abou 3.4% annually. 13

20 Table 3. Hedge Fund Performance Before and During 13F Porfolio Disclosure This able repors he resuls of univariae ess on hedge fund performance before and during 13F porfolio disclosure. The saisical significance on he difference in performance is obained using paired -ess. Before 13F Filing During 13F Filing Difference n mean sd n mean sd Paired- es Raw Reurns (%) **** **** **** Marke Model Alpha (%) **** **** **** Bea-Marke **** **** *** Adj R-squared *** Fama-French 3 Facor Model Alpha (%) **** *** **** Bea-Marke **** **** *** Bea-SMB **** **** *** Bea-HML **** **** R-squared (Adj) *** Fama-French 4 Facor Model Alpha (%) *** **** Bea-Marke **** **** *** Bea-SMB **** **** ** Bea-HML **** **** Bea-Momenum R-squared (Adj) *** Fung and Hsieh 7 Facor Model Alpha (%) **** Bea-Bond Trend-Following * Bea-Currency Trend-Following ** Bea-Commodiy Trend-Following ** Bea-S&P **** **** Bea-SC-LC **** *** Bea-10-year Treasury Yield Bea-Credi Spread ** **** R-squared (Adj) *** 14

21 Table 4. Calendar Time Porfolios This able repors raw and risk-adjused reurns for a calendar ime porfolio ha longs disclosed funds and shors non-disclosed funds. In monh, a fund family is classified as disclosed if i files 13F in monh -1, oherwise i is classified as non-disclosed. 15

22 16 (1) (2) (3) (4) (5) Raw Reurn Marke Model FF-3Facor Model Carhar 4 Facor Model Fung-Hsieh 7 Facor Model Consan ** *** ** ** ** (-2.34) (-2.74) (-2.24) (-2.23) (-2.12) Marke 0.147**** *** ** (3.68) (2.71) (2.27) SMB (-0.29) (-0.35) HML **** **** (-4.66) (-4.56) Momenum (0.31) Bond Trend-Following (0.48) Currency Trend-Following (0.51) Commodiy Trend-Following (0.49) S&P *** (3.24) SC_LC (-0.52) Credi Spread (-0.52) 10-year Treasury Yield (-0.04) Observaions Adjused R-squared

23 Regression Disconinuiy (RD) Design The disconinuous change in he disclosure regime ha akes effec when invesmen companies' asses cross he $100 million hreshold allows me o idenify an effec of porfolio disclosure on hedge fund performance ha is purged of poenial endogeneiy problems. I employ a regression disconinuiy design in which he idenifying assumpion is ha he funcion ha relaes fund size o performance does no have precisely he same jumps as he funcion ha relaes fund size o disclosure. This procedure is valid even if unobserved facors such as fund manager skill ha affec performance are funcionally relaed o fund size. I narrow he sample o a small neighborhood around he $100 million hreshold and use model below o deec wheher here is a jump in performance when he disclosure requiremen changes. Specifically, I keep he fund-year observaions ha have a lagged fund size greaer han $70 million and less han $130 million. The resuls are robus o various widh of he neighborhood such as $90 million o $110 million and $80 million o $120 million. Performanc e, 2 = α + β Disclosure + γ 1Size 1 + γ 2Size 1 + ε i where Performance i; is a performance measure for fund family i in year. Performance i; is equal o 0 if year is before invesmen company i begins filing form 13F, and is equal o 1 if year is during he period ha invesmen company files form 13F. The coefficien on Disclosure i; capures wheher here is a jump in performance when he disclosure code changes from 0 o 1. Boh linear and quadraic erms of lagged asses under managemen are included in he model o 17

24 conrol for he effec of fund size on fund performance. Using samples only in he small neighborhood of $100 million also allows he resuls o be less dependen on he model specificaions, such as he quadraic relaion beween size and performance. As shown in columns (1), (3), (5), (7), and (9) in Panel A of Table 5, he coefficien on Disclosure i; is negaive and saisically significan across all hree performance measures. There is a drop in performance ha occurs in he form of a disconinuous jump ha canno be explained by coninuous changes in asses under managemen when he regulaory regime swiches from non-disclosure o disclosure. There are also funds in TASS ha have crossed he hreshold of $100 million bu were no required o file 13F because some or all of heir asses are no 13(f) securiies, are shor posiions, conain fewer han 10,000 shares, or have a marke value of less han $200,000. These non-13f filing funds provide he opporuniy for a conrol es. If he drop in performance is due o 13F disclosure, we should no observe a disconinuous drop in performance for his conrol group when hey cross he $100 million hreshold. As shown in columns (2), (4), (6), (8), and (10) in Panel A of Table 5, he coefficien on variable Disclosure i; is no significanly differen from zero, which suggess ha here is no drop in performance for funds ha crossed he $100 million hreshold bu did no file 13F disclosure. I also invesigae wheher here is a disconinuous drop when funds cross 18

25 oher hresholds (e.g., $70 million or $130 million). If he drop in performance when funds crossed he $100 million hreshold is due o 13F disclosure, we should no observe drops in performance when funds cross oher hresholds. As shown in Panel B of Table 5, he coefficien on Dummy i; is no saisically differen from zero when using he hreshold of $70 million or $130 million. These resuls confirm ha he drop in performance following disclosure is no due o a change in size or mean reversion in reurns. To address he concern ha he yearly performance measure alphas esimaed wih welve monhly reurn observaions may no be very reliable due o he limiaion of he sample size, I scaled each alpha esimae by is sandard error and hen run ess idenical o hose repored in Panels A and B and he resuls are repored in Panels C and D. By scaling he alpha esimae by is sandard error, I give greaer weigh o he alpha esimaes which are relaive more precise. As shown in Panel C and Panel D, he resuls are qualiaively similar. 19

26 Table 5. Regression Disconinuiy Design 20 This able repors he resuls of regression disconinuiy (RD) design wih samples narrowed o a small neighborhood around he hreshold of $100 million (δ is chosen o represen $30 million). In Panel A, funds ha filed 13F during he sample period are included in he analysis in model (1), (3), (5), (7), and (9). Performance is a performance measure for fund family i in year. I use five differen performance measures in my analysis: raw reurns, marke model alpha, Fama-French hree-facor model alpha, Carhar four-facor model alpha, and Fung-Hsieh seven-facor model alpha. All are calculaed using monhly fund reurns repored in TASS. Disclosure is equal o 0 if year is before he invesmen company i sars o file form 13F, and is equal o 1 if year is during he period ha invesmen company files form 13F. 2 Performanc e = α + β Disclosure + γ 1 Size 1 + γ 2 Size 1 + ε Sample: $100million δ <=Size -1 <= $100million + δ Funds ha never filed 13F during he sample period are used as a conrol group for he analysis, and resuls are repored in model (2), (4), (6), (8), and (10). Dummy is equal o 1 if he size of invesmen company i in year -1 is equal or greaer han $100 million, and is equal o 0 oherwise. Performanc e Dummy Size Size, 1, if Sizei, 1 >= $100 million Dummy = 0, if Sizei, 1 < $100 million Sample: $100million δ <=Size -1 <= $100million + δ 2 = α + β + γ γ ε i In Panel B, oher hresholds, including $70 million and $130 million, are chosen for he analysis. Specifically, Dummy is equal o 1 if he size of invesmen company i in year -1 is equal or greaer han $70 (or $130) million, and is equal o 0 oherwise. 2 Performanc e = α + β Dummy + γ 1 Size 1 + γ 2 Size 1 + ε 1, if Sizei, 1 >= $70 ( or $130) million Dummy = 0, if Sizei, 1 < $70 ( or $130) million Sample: $70 (or $130 million) δ <=Size -1 <= $70 (or $130 million) + δ

27 21 In Panel C, he performance measure - alpha for invesmen company i in year is scaled by he sandard error of he alpha esimae. These error-scaled alphas are hen used in he regression disconinuiy (RD) design ha is idenical o he ess in Panel A. Similarly, he alphas used in Panel D are scaled by heir sandard errors and he ess are idenical o hose in Panel B.

28 Panel A. Funds ha Did and Did No File 13F When Crossing he Threshold of $100 Million Raw Reurn Marke Model Alpha FF 3-Facor Alpha Carhar 4-Facor Alpha Fung-Hsieh 7-Facor Alpha Tes Conrol Tes Conrol Tes Conrol Tes Conrol Tes Conrol Group Group Group Group Group Group Group Group Group Group (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Disclosure ** ** ** * /Dummy (-1.59) (-0.70) (-2.26) (0.71) (-2.78) (-0.62) (-2.58) (-0.73) (-1.81) (-1.16) Lagged Size (10-10 ) (0.56) (1.51) (-0.42) (1.02) (-0.93) (0.28) (-1.02) (0.13) (-1.40) (0.54) Lagged Size (10-18 ) (-0.60) (-1.47) (0.35) (-1.03) (0.83) (-0.07) (0.89) (0.09) (1.23) (-0.39) 22 Consan (-0.39) (-1.23) (0.66) (-0.78) (1.21) (-0.29) (1.30) (-0.19) (1.68) (-0.53) Observaions R-squared *, **, ***, and **** denoe saisical significance a 10, 5, 1, and 0.1 percen level, respecively.

29 Panel B. Oher Thresholds ($70 million and $130 million) Raw Reurn Marke Model Alpha FF 3-Facor Alpha Carhar 4-Facor Alpha Fung-Hsieh 7-Facor Alpha $70 M $130 M $70M $130M $70M $130M $70M $130M $70M $130M Disclosure /Dummy (0.01) (1.25) (0.33) (0.92) (-0.01) (-0.35) (0.13) (-0.88) (-0.43) (0.57) Lagged Size (10-10 ) (-1.23) (-0.17) (-1.63) (0.63) (-1.30) (1.03) (-1.18) (0.75) (-0.24) (0.60) Lagged Size (10-18 ) (1.26) (-0.04) (1.36) (-0.75) (1.29) (-0.94) (1.09) (-0.62) (0.42) (-0.60) Consan ** *** (2.89) (0.45) (3.10) (-0.45) (1.57) (-1.04) (1.35) (-0.79) (0.37) (-0.51) 23 Observaions R-squared *, **, ***, and **** denoe saisical significance a 10, 5, 1, and 0.1 percen level, respecively.

30 Panel C. Funds ha Did and Did No File 13F When Crossing he Threshold of $100 Million (Error-Scaled) Raw Reurn Marke Model Alpha FF 3-Facor Alpha Carhar 4-Facor Alpha Fung-Hsieh 7-Facor Alpha Tes Conrol Tes Conrol Tes Conrol Tes Conrol Tes Conrol 24 Group Group Group Group Group Group Group Group Group Group (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Disclosure ** *** 0.371* *** *** *** /Dummy (-2.46) (1.32) (-3.53) (1.87) (-3.57) (1.04) (-3.84) (1.06) (-3.37) (0.55) Lagged Size (10-8 ) (-0.14) (-0.90) (-0.74) (-0.93) (-1.44) (-1.00) (-1.58) (-1.08) (-0.38) (-1.12) Lagged Size (10-16 ) (0.14) (0.84) (0.77) (0.84) (1.53) (0.90) (1.62) (0.95) (0.38) (1.14) Consan * (0.51) (1.22) (1.04) (1.22) (1.67) (1.30) (1.87) (1.37) (0.67) (1.17) Observaions R-squared *, **, ***, and **** denoe saisical significance a 10, 5, 1, and 0.1 percen level, respecively.

31 Panel D. Oher Thresholds ($70 million and $130 million) (Error-Scaled) Raw Reurn Marke Model Alpha FF 3- Facor Alpha Carhar 4-Facor Alpha Fung-Hsieh 7-Facor Alpha $70 M $130 M $70 M $130 M $70 M $130 M $70 M $130 M $70 M $130 M Disclosure /Dummy (1.06) (-0.20) (1.72) (-0.77) (1.61) (-0.94) (1.34) (-0.89) (0.73) (-0.45) Lagged Size (10-8 ) (0.96) (0.64) (0.65) (0.58) (0.75) (0.60) (0.63) (0.45) (1.18) (0.38) Lagged Size (10-16 ) (-1.00) (-0.58) (-0.79) (-0.50) (-0.90) (-0.53) (-0.79) (-0.36) (-1.13) (-0.29) Consan (0.05) (-0.50) (0.08) (-0.49) (-0.19) (-0.51) (-0.17) (-0.40) (-0.83) (-0.33) 25 Observaions R-squared *, **, ***, and **** denoe saisical significance a 10, 5, 1, and 0.1 percen level, respecively.

32 Disconinuous Change in Performance When Funds Sar o File 13 The univariae ess in Table 3 show ha fund performance is poorer in he disclosure periods han in he non-disclosure periods. If he decreased performance is due o porfolio disclosure, he change in fund performance should no occur slowly over ime, bu in he form of a sudden drop immediaely afer funds sar o file 13F disclosure. In his secion, I use he following regression o es wheher he change in performance occurs as soon as funds begin filing 13F: Change in Performance β 4 4 h Year + β 5 5 h+ = α + β γ Change in Size s Year + β nd Year + β 3 + γ Change in Size 3 rd 2 1 Year + Change in Performance i; is he change in performance of fund family i in year from year -1. The variable 1 s Year i; is equal o 1 if i is he firs year in which he fund family i sars o file 13F. I also include dummy variables for 2 nd, 3 rd, and 4 h year and for 5 h year or laer. The conrol variables include lagged change in linear and quadraic erm of asses under managemen. All TASS funds excep funds ha repor in a currency oher han U.S. dollars are included in he analysis, and back-filled daa are removed. Fund family and year fixed effecs are included in he model and errors are clusered. As shown in Panel A Table 6, he coefficiens on variable 1 s Year i; are negaive and saisically significan across all five performance measures. For example, he coefficien on variable 1 s Year i; in model (1) is , which suggess ha he drop in raw reurns in he firs year afer funds sar o file 13F is 0.52% monhly (or 6.3% annually). However, he coefficiens on 2 nd, 3 rd, 4 h, and 26

33 5 h+ year dummies are no negaively saisically significan. The resuls sugges ha he decrease in performance occurs in he firs year afer funds sar o file 13F, which suppors he argumen ha he drop in performance is due o 13F disclosure. The coefficien on lagged change in asses under managemen is negaive and saisically significan, which suggess diminishing reurns o scale. The coefficien on lagged change in squared size is posiive, which suggess ha he relaion beween performance and size is concave. This disconinuous change in performance along he ime dimension again lends srong suppor for he argumen ha porfolio disclosure harms fund performance. Panel B repors he resuls where he alpha esimaes are scaled by is sandard error before being used in he es. As discussed in he previous session, his reamen is o address he concern ha he sample size for esimaing alpha in each year is limied (12 monhly observaions). The resuls remain unchanged afer applying his reamen. 27

34 Table 6. Change in Performance When Funds Sar o File 13F This able repors he resuls of he following regression model: Change in Performance β 4 4 h Year + β 5 5 h+ = α + β γ Change in Size s Year + β nd Year + β 3 + γ Change in Size 3 rd 2 1 Year + Change in Performance is he change in performance of fund family i in year from year -1. The variable 1 s Year is equal o 1 if i is he firs year since fund family i sar o file 13F. I also include dummy variables for 2 nd, 3 rd, 4 h, and 5 h year or laer. The conrol variables include lagged change in linear and quadraic erm of asses under managemen. All invesmen companies ha repor o he TASS daase are included in he analysis. Fund family and year fixed effecs are included in he model and errors are clusered. The sample period is from June 1990 o February 2010 afer removing he backfilled daa

35 Panel A. Change in Performance When Funds Sar o File 13F Change in Alpha Raw Reurn Marke Model FF3 Facor Carhar 4 Facor Fung-Hsieh 7 Facor (1) (2) (3) (4) (5) 1 s Year * * *** *** * (-2.02) (-1.83) (-3.63) (-3.63) (-1.80) 2 nd Year (-0.75) (-1.17) (-0.74) (-0.69) (-0.99) 3 rd Year (-0.87) (-0.99) (-0.93) (-0.38) (-0.07) 29 4 h Year (-0.76) (-0.52) (-0.63) (-0.29) (-0.05) 5 h+ Year (-0.65) (-0.91) (-1.53) (-1.19) (-0.75) Lag Change in Size -1.36**** **** **** **** *** (x ) (-6.16) (-9.66) (-7.27) (-6.69) (-3.12) Lag Change in Size **** 5.64**** 4.89**** 4.50**** 3.55*** (x ) (6.80) (8.89) (5.86) (5.55) (4.05) Consan (1.28) (0.74) (0.24) (1.05) (1.50) Observaions R-squared *, **, ***, and **** denoe saisical significance a 10, 5, 1, and 0.1 percen level, respecively. 29

36 Panel B. Change in Performance When Funds Sar o File 13F (Error Correced) Change in Alpha Raw Reurn Marke Model FF3 Facor Carhar 4 Facor Fung-Hsieh 7 Facor (1) (2) (3) (4) (5) 30 1 s Year * ** *** **** * (-1.79) (-2.57) (-3.71) (-4.42) (-2.03) 2 nd Year (-1.72) (-1.39) (-1.65) (-1.18) (-1.04) 3 rd Year (0.26) (-0.02) (0.35) (0.43) (0.50) 4 h Year (-0.92) (-0.56) (-0.70) (-0.32) (0.08) 5 h+ Year (-1.27) (-0.94) (-1.44) (-1.33) (-0.80) Lag Change in Size -1.35**** -1.08**** -8.59*** -7.25*** -3.45** (x ) (-5.56) (-5.91) (-3.98) (-4.05) (-2.90) Lag Change in Size **** 6.59**** 5.21*** 4.27*** 2.48*** (x ) (5.18) (5.67) (3.72) (3.65) (3.76) Consan (1.07) (0.84) (0.41) (1.14) (1.03) Observaions R-squared *, **, ***, and **** denoe saisical significance a 10, 5, 1, and 0.1 percen level, respecively. 30

37 Illiquidiy and he Impac of Porfolio Disclosure on Performance The findings in he previous secions sugges ha 13F porfolio disclosure leads o lower hedge fund performance. In his secion, I examine he hypohesis ha he cos of porfolio disclosure should be greaer for funds ha hold illiquid asses. In general, rades in illiquid securiies resul in larger price impacs han rades in liquid securiies. In order o reduce he ransacion cos caused by price impac, fund managers end o employ sequenial rading sraegies o accumulae or dispose of an illiquid posiion. However, he longer i akes o accumulae or dispose of a posiion, he higher he likelihood and he greaer he cos of being aken advanage of by fronrunners and free riders. Consisen wih hese argumens, I find ha he drop in performance is more dramaic for funds wih illiquid holdings. I use he following regression model o es wheher he cos of disclosure is greaer for funds wih illiquid holdings: Performance + β log Size 1 1 = α + β Disclosure + β (log Size) γ Illiquidiy 1 i Disclosure + γ Illiquidiy 2 Illiquidiy i is he average Amihud (2002) illiquidiy measure calculaed using he holdings disclosed in form 13F for fund family i over all disclosing quarers. The Amihud (2002) illiquidiy measure is he average daily illiquidiy during he quarer preceding he 13F filing quarer (where daily illiquidiy is calculaed as he absolue reurn divided by he dollar rading volume on ha day): Illiq Q = 1 N N T 1 re vol prc 31

38 where Illiq Q is quarerly illiquidiy, N is he number of days in he quarer, and re, vol, and prc are he reurn, rading volume, and he price on day, respecively. As shown in Table 7, he coefficiens on he ineracion erm beween Illiquidiy i and Disclosure i; are negaive across all five models and are saisically significan excep in models (1). These resuls sugges ha porfolio disclosure is more cosly for funds ha have more illiquid holdings. The Impac of 13F Disclosure on Correlaions of Hedge Fund Reurns If porfolio disclosure reveals rade secres and faciliaes free-riding aciviies, we should expec o observe an increase in correlaions beween he reurns of fund i and he reurns of oher hedge funds afer fund i sars o file 13F disclosure. To measure he correlaions beween he reurns of fund i and oher hedge funds ha are in he same invesmen syle in year, I regress he monhly reurns of fund i on he value-weighed reurns in year of all hedge funds ha are in he same invesmen syle. The R-squared obained from his regression describes how much of he reurn variaion for fund i can be explained by he index reurns for all hedge funds ha are in he same invesmen syle and is used as a measure of correlaions beween he fund i and oher hedge funds ha are in he same invesmen syle. The regression model in column (1) and (2) of Table 8 is: 2 R = α + β1 Disclosure + ε Where Disclosure i; is a dummy variable ha is equal o 1 if fund i files Form 13F in year, oherwise 0. Advisor fixed effecs are included in he models. The 32

39 models in column (3) and (4) add conrol variables including lagged size and lagged quadraic erms of size. As shown in Table 8, he coefficien on variable Disclosure i; is posiive and saisically significan, which indicaes ha here is an increase in reurn correlaions beween fund i and oher hedge funds ha are in he same invesmen syle afer fund i sars o file he 13F disclosure. The increase in R 2 is abou 3%, which is also economically significan. These resuls provide srong evidence ha here is an increase in reurn correlaions beween disclosing funds and oher hedge funds when funds sar o file 13F. These findings suppor he argumen ha porfolio disclosure reveals rade secres and faciliaes free-riding aciviies. 33

40 Table 7. Illiquidiy and he Impac of Porfolio Disclosure on Hedge Fund Performance This able repors he resuls of he following regression model: Performance 2 = α + β Disclosure + γ 1 Illiquidiyi Disclosure + γ 2 Illiquidiy + β1 log Size 1 + β2 (log Size) 1 Illiquidiy is he Amihud (2002) illiquidiy measure calculaed based on he holdings disclosed in 13F. The Amihud (2002) illiquidiy measure is calculaed as he average daily illiquidiy during he quarer preceding he 13F filing quarer (where daily illiquidiy is calculaed as he absolue reurn divided by he dollar rading volume on ha day): 34 Illiq Q = 1 N N = 1 vol re prc where Illiq Q is quarerly illiquidiy, N is he number of days in he quarer, re, vol, and prc are he daily reurn, rading volume, and he price on day, respecively.

41 35 Raw Alpha (Marke Model) Alpha (FF3 Facors) Alpha (Carhar 4 Facors) Alpha (FT7 Facors) (1) (2) (3) (4) (5) Disclosure * ** ** * (-1.01) (-2.22) (-2.85) (-2.51) (-2.23) Lagged log Size *** (-4.07) (-1.72) (-0.48) (-0.32) (-0.14) Lagged (log Size) *** (4.21) (1.86) (0.51) (0.31) (0.30) Disclosure x Illiquidiy * ** * * (-0.97) (-2.07) (-2.72) (-2.34) (-2.17) Illiquidiy ** ** * (1.36) (1.86) (3.43) (2.68) (2.27) Consan *** ** (4.02) (2.55) (1.38) (1.28) (1.00) Observaions R-squared *, **, ***, and **** denoe saisical significance a 10, 5, 1, and 0.1 percen level, respecively.

42 Table 8. The Impac of 13F Disclosure on he Correlaions of Hedge Fund Reurns The regression model in he able below is: 2 R = α + β1 Disclosure + ε 2 where R is he R-squared from regressing monhly reurns of fund family i on he value-weighed reurns of hedge funds in he same syle in year. Disclosure is a dummy variable which is equal o 1 if fund family i files Form 13F in year, oherwise 0. Fund fixed effecs are included in he model. 36 R-squared (1) (2) (3) (4) Disclosure *** * ** ** (2.77) (1.72) (2.48) (2.35) Lag log Size ** ** (-2.26) (-2.05) Lag log Size **** *** (3.66) (2.93) Consan 0.419**** 0.441**** 0.553*** 0.914*** (143.76) (4.11) (2.85) (3.19) Year Dummy No Yes No Yes Observaions R-squared *, **, ***, and **** denoe saisical significance a 10, 5, 1, and 0.1 percen level, respecively.

43 The Impac of 13F Disclosure on Compensaion Schemes Since porfolio disclosure allows more monioring of fund aciviy, i inroduces anoher mechanism o conrol he agency problem. Though agency heories have rich implicaions for how monioring may inerac wih he fund manager's compensaion incenives, he impac of disclosure on opimal compensaion is sill a rarely explored empirical quesion in he lieraure. The ypical compensaion in he hedge fund indusry includes a managemen fee ha is a fixed percenage of asses and an incenive fee ha is a percenage of he profi when a fund reurn is posiive or exceeds a high-waer mark. This incenive fee aligns he ineres of managers wih ha of invesors. However, such opion-like fee srucures also provide managers wih an incenive o ake invesmen risks ha exceed invesors' desired risk level. Invesors wih informaion abou a porfolio's holdings are beer able o assess he risk hey are exposed o, a circumsance ha disciplines risk aking by fund managers. Porfolio disclosure hereby reduces he cos of opion-like incenive fees and allows for higher incenive fees in he opimal compensaion srucure. Though TASS only provides a snapsho of he fee srucure of each reporing fund, he incepion dae of each fund ells wheher he fee srucure was se before or during he period ha he fund family filed 13F disclosure. Because fund fees are se a he ime of he incepion of he fund, I compare he fees of he funds launched during he period ha heir fund families file 13F wih he fees of he funds launched before 13F disclosure. In he previous secion, he analysis of 37

44 he impac of porfolio disclosure on hedge fund performance is conduced a he fund family level. In his secion, he analysis is conduced a he fund level. Summary Saisics on Hedge Fund Compensaion Schemes Table 9 provides summary saisics on hedge fund compensaion schemes, including incenive fees, managemen fees, and wheher funds use a high-waer mark. Funds of funds are excluded from he sample because heir fee srucures are differen. Funds ha launched afer heir fund family sopped filing form 13F are also excluded. Panel A repors and compares compensaion schemes for non- 13F filers and 13F filers. Similar o he definiion earlier, non-13f filers refers o funds ha never filed 13F during he sample period and 13F filers refers o funds ha belong o a fund family ha filed a 13F a leas once. As shown in Panel A of Table 8, he average incenive fee for 13F filers is 18.92% and for non-13f filers is 18.16%. Thus, he incenive fees for 13F filers are 0.76% greaer han for non- 13F filers, and he difference is saisically significan. The average managemen fee for boh 13F filers and non-13f filers is 1.50%. The percenage of funds ha use a high-waer mark for 13F filers is 67% and for non-13f filers is 84%. The difference is 17% and is saisically significan. Panel B repors and compares he compensaion scheme for funds ha launched before heir fund families sared o file 13F and for funds ha launched during he period ha heir fund families filed 13F. The univariae es shows ha funds ha launched during he 13F filing have higher incenive fees (0.56%, bu no saisically significan) and use a highwaer mark more frequenly. However, such univariae ess canno conrol for 38

45 oher facors, such as ime rends, ha migh have affeced he compensaion scheme. I show he resuls of mulivariae ess in he nex secion. The Impac of 13F Disclosure on Compensaion Schemes I use he following mulivariae regression o capure he effec of disclosure on incenive fees: Incenive Fee i = + β Disclosurei + γ 1 FamilyAsses + γ 2 α FamilyAge + ε i where he dependen variable is incenive fees as a percenage erm. The variable Disclosure i is equal o 1 if he fee srucure is se during he period ha a fund family files 13F, and is equal o 0 if he fee srucure is se in he period before he fund family sars o file 13F. The conrol variables include he fund family's asses under managemen and family age (in log form) a he fund's incepion dae, and a dummy variable ha indicaes wheher he fund family has ever filed a 13F repor. Year, fund caegory, and fund family fixed effecs are included in he models and errors are clusered. As shown in columns (1) o (5) of Table 10, he coefficiens on Disclosure i are posiive and saisically significan. The incenive fees are on average abou 1 % higher when a fund family files 13F porfolio disclosure. In unrepored ables, I find ha here is no change in he managemen fee or he use of a high-waer mark afer funds file he 13F disclosure. Overall, hese resuls suppor he hypohesis ha porfolio monioring reduces risk-aking by fund managers and allows for higher convexiy in he opimal compensaion srucure. 39