Investing in talents mager characteristics and hegde fund performance

However, in contrast to the results for mutual funds, wefind a rather symmetric relation between hedge fund flows and past performance, and that hedgefund flows do not have a significant

Investing in Talents: Manager Characteristics and

Hedge Fund Performances

Haitao Lia, Xiaoyan Zhangb, and Rui Zhaoc

Investing in Talents: Manager Characteristics and Hedge Fund Performances

AbstractUsing a large sample of hedge fund manager characteristics, we provide one of the first compre-hensive studies on the impact of manager characteristics, such as education and career concern, onhedge fund performances We document differential ability among hedge fund managers in gen-erating risk-adjusted returns and flow-chasing-return behaviors among hedge fund investors Inparticular, we find that managers from higher-SAT undergraduate institutes tend to have higherraw and risk-adjusted returns, more inflows, and take less risks Our results provide supportingevidence to some of the assumptions and implications of the rational theory of active portfoliomanagement of Berk and Green (2004) However, in contrast to the results for mutual funds, wefind a rather symmetric relation between hedge fund flows and past performance, and that hedgefund flows do not have a significant negative impact on future performance

JEL: G23, G11, G12

Keywords: hedge fund performance, manager characteristics, hedge fund flows

An investment in a hedge fund is really an investment in a manager and the specialized talent

he possesses to capture profits from a unique strategy.–– Sanford J Grossman, The Wall StreetJournal, September 29, 2005

Hedge funds have experienced tremendous growth in the past decade According to the SECand various hedge fund research companies, the amount of assets under management by hedgefunds has grown from about $15 billion in 1990 to about $1 trillion by the end of 2004, andthe number of existing hedge funds is about 7,000 to 8,000 Some industry experts even predictthat hedge fund assets could exceed $3.2 trillion globally by 2009 As a result, hedge funds haveattracted enormous attention from a wide range of market participants and academics in recentyears

Hedge funds differ from mutual funds in the ways they operate and how their managers arecompensated For example, hedge funds are not subject to the same level of regulation as mutualfunds and thus enjoy greater flexibility in their investment strategies As a result, hedge fundsfrequently use short selling, leverage, and derivatives, strategies rarely used by mutual funds, toenhance returns and/or reduce risk While mutual funds charge a management fee proportional toassets under management (usually 1-2%), most hedge funds charge an incentive fee, typically 15%

to 20% of profits, in addition to a fixed 1-2% management fee Moreover, hedge fund managersoften invest a significant portion of their personal wealth in the funds they manage; and manyfunds have a high watermark provision, which requires managers to recoup previous losses beforereceiving incentive fees

Hedge funds also differ from mutual funds in the economic functions they perform in the omy As pointed out by Sanford J Grossman (2005) in a recent Wall Street Journal commentary,while mutual funds enable small investors to pool their money and invest in diversified portfolios,

econ-“a hedge fund is a vehicle for acquiring the specialized talents of its manager.” Grossman observesthat, “Hedge funds are typically managed by an entrepreneur, and hedge fund returns are theoutcome of an entrepreneur activity.” As a result, Grossman emphasizes that a “fund’s returnwill be no better than its management and the economic environment in which it produces itsproduct An investor should understand the product being produced and the manager producingit.” Grossman’s observation suggests that the performance of a hedge fund depends crucially onboth the investment strategies it follows and the talents of its manager(s) in implementing suchstrategies

Though great progress has been made in understanding the risk and return properties of manyhedge fund strategies,1 only limited analysis has been done on the impact of manager talents onhedge fund performances in the literature Just like any entrepreneur activity, it is entirely possible1

See, for example, the interesting works of Agarwal and Naik (2004), Fung and Hsieh (1997, 2001), and Mitchell and Pulvino (2001), among others.

that some hedge fund managers are better than others in making investment decisions Given thebillions of dollars poured into hedge funds from pension funds, endowments, and other institutionalinvestors each year, identifying manager characteristics that lead to superior performance could bevery helpful to potential investors in selecting hedge fund managers and also could have profoundwelfare implications.

In addition to the practical value of identifying superior managers, understanding the impact

of manager talents on hedge fund performances also provides a way of testing some of the tions and implications of the rational theory of active portfolio management of Berk and Green(2004).2 For example, one important assumption of Berk and Green (2004) is that mutual fundmanagers have differential ability in delivering positive risk-adjusted returns However, if the the-ory of Berk and Green (2004) is true, then it could be difficult to identify cross-sectional differences

assump-in risk-adjusted returns assump-in equilibrium usassump-ing mutual fund data, because most mutual funds mighthave increased their sizes to the extent that their risk-adjusted returns have disappeared More-over, due to established investment process and team-oriented approach to portfolio management

in many mutual fund families, the impact of individual managers on mutual fund performances

is likely to be small as well Consistent with this view, Chevalier and Ellison (1999a) find thatalthough mutual fund managers from higher-SAT institutes tend to have higher raw returns, theirresults become much less significant for risk-adjusted returns

In contrast, the unique structure of hedge funds suggests that manager talents might be moreimportant for hedge fund performances Since a significant part of hedge fund compensationcomes from incentive fees, hedge fund managers may not want to grow their funds to the extentthat all risk-adjusted returns disappear In addition, many hedge funds have a high watermarkprovision, and many hedge fund managers have personal wealth invested in their funds As aresult, inferior hedge fund returns could be really costly for these managers Therefore, even inequilibrium there might be an optimal fund size at which abnormal returns still exist In addition,the entrepreneur nature of hedge fund operations suggests that hedge fund performance shoulddepend more significantly on individual managers Therefore, hedge fund data provide a uniqueopportunity for testing the theory of Berk and Green (2004)

In this paper, we provide a comprehensive empirical analysis of the impact of manager acteristics on hedge fund performances We conjecture that everything else equal, a manager who

char-is more talented and more devoted to hchar-is/her job char-is more likely to have better performance Weuse intelligence and education as proxies for manager talents We use manager career concern as

2 Berk and Green’s (2004) model combines three elements: competitive provision of capital by investors to mutual funds, differential ability to generate high average returns across managers but decreasing returns to scale

in deploying these abilities, and learning about managerial ability from past returns The theory predicts that mutual fund managers increase the size of their funds, and their own compensation, to the point at which expected returns to investors do not outperform passive benchmarks in equilibrium.

a proxy for manager job commitments The rationale is that a manager who is under pressure

to establish his/her career at an early stage might be willing to put in more effort than a moreestablished manager

We first construct probably the most comprehensive dataset on manager characteristics based

on more than 4,000 hedge funds covered by TASS between 1994 and 2003 Boyson (2003, 2004)studies hedge fund performance and manager career concerns using a much smaller sample ofabout 200 funds up to 2000 In contrast, our dataset covers a wide range of information on per-sonal, educational, and professional backgrounds of managers of 1,002 hedge funds up to 2003.Specifically, we collect information on the following six characteristics of the lead manager of eachfund if such information is available: the composite SAT score for the manager’s undergraduateinstitute (SAT), whether the manager has a CPA or CFA, whether the manager has an MBA de-gree, the total number of years of working (WORK), the number of years of working at the specifichedge fund (TENURE), and the manager’s age (AGE) Broadly speaking, the six characteristicscan be divided into two groups: SAT, CFA/CPA and MBA dummies represent intelligence andeducation; WORK, TENURE, and AGE could represent working experience and career concern

We also conduct a careful analysis on risk adjustments for hedge fund returns to obtain hedgefund abnormal performance Many studies have shown that due to the dynamic trading strategiesand derivatives used by hedge funds, traditional linear asset pricing models could give misleadingresults on hedge fund performance Given that there are no well-established risk-adjustmentmethods for hedge fund returns, we choose a wide variety of models to ensure the robustness ofour results Specifically, in addition to the traditional Fama and French (1993) (hereafter FF)three-factor model, to capture the nonlinearity in hedge fund returns, we also consider a widevariety of models that include returns on various hedge fund indices and options as factors Inparticular, we consider the model of Agarwal and Naik (2004) and the seven-factor model firstproposed by Fung and Hsieh (2004) and used recently by Fung, Hsieh, Naik, and Ramadorai(2006) (hereafter FHNR) As a further robustness check, we consider two specific hedge fundstrategies whose risk and return properties have been carefully examined and thus are reasonablywell understood in the literature These are the trend following strategy studied by Fung andHsieh (2001) and the risk arbitrage strategy studied by Mitchell and Pulvino (2001)

Based on the new dataset on manager characteristics and various risk-adjustment methods, wedocument a strong impact of manager education on different aspects of hedge fund performances,such as fund risk-taking behaviors, raw and risk-adjusted returns, and fund flows Specifically,

we find that managers from higher-SAT institutes tend to take less (overall, systematic, andidiosyncratic) risks and have higher raw and risk-adjusted returns In our analysis, risk-adjustedreturns include both alpha and appraisal ratio (the ratio between alpha and residual volatility)

We also find that managers from higher-SAT institutes tend to attract more capital inflows On

the other hand, we find some weak evidence that managers with longer years of working tend

to have lower raw and risk-adjusted returns and take less risks These results are very robust tothe different risk-adjustment benchmarks, sample periods, and types of funds (funds of funds vs.regular hedge funds) we consider

Although we document differential ability among hedge fund managers in generating adjusted returns and flow-chasing-return behaviors among hedge fund investors, we find mixedresults in our tests of other implications of Berk and Green’s theory For example, unlike theconvex relation between flows and lagged returns documented for mutual funds, we find thathedge fund flows react to lagged returns rather symmetrically We also find a significant androbust negative relation between hedge fund flows and both fund age and lagged fund size Thissuggests that there might be an optimal fund size beyond which hedge fund managers start totake less inflows Finally, in contrast to the results for mutual funds, we do not find a significantnegative impact of current fund flows on future fund performances for hedge funds

risk-Our paper contributes to the fast-growing literature on hedge funds by providing (i) one

of the first systematic studies on the impact of manager characteristics on the cross-sectionaldifferences in hedge fund performances and (ii) an empirical test of Berk and Green’s (2004)theory using hedge fund data Our paper also complements and extends FHNR (2006), the firststudy that tests Berk and Green’s (2004) theory using hedge fund data.3 While both FHNR(2006) and our paper show that some hedge fund managers are indeed better than others, ourstudy traces superior hedge fund performances to important manager characteristics, such aseducation and career concern Therefore, our paper provides an economic explanation for theexistence of superior performances as well as a guidance on how to identify superior hedge fundmanagers based on manager characteristics Our results on flow-return relation also are broadlyconsistent with that of FHNR (2006) While FHNR (2006) show that fund flows negatively affectthe transition probability of have-alpha funds to remain in have-alpha category, the effect of flows

on future risk-adjusted returns is not statistically significant Collectively, the results of our paperand FHNR (2006) suggest that the basic mechanisms of Berk and Green’s (2004) model are also

at work in the hedge fund industry However, because of the unique compensation structure ofhedge funds, hedge fund managers do not have the same incentives as mutual fund managers ingrowing the size of their funds Therefore, the negative impact of fund flows on future returnsfor hedge funds may not be as strong as that for mutual funds, and hedge funds may still exhibitpositive abnormal returns even in equilibrium.4 Our results strongly suggest that hedge funds are

3 Using data on funds of funds, FHNR (2006) show that some hedge fund managers are able to deliver better alphas than others They further show that the alpha producing funds of funds (denoted as have-alpha funds) experience greater and steadier capital inflows than the other funds that fail to produce alphas (denoted as beta- only funds).

4 This view is also consistent with the findings of Kosowski, Naik, and Teo (2007) Using powerful bootstrap

very different from mutual funds, and a manager’s talents and motivations should be importantconsiderations in selecting hedge fund managers.

The remainder of the paper proceeds as follows In section I, we introduce our data on hedgefund returns and manager characteristics In section II, we introduce a wide variety of risk-adjustment benchmarks for hedge fund returns In section III, we examine the relation betweendifferent aspects of hedge fund performance and manager education/career concerns In Section

IV, we specialize our analysis to two special hedge fund strategies whose risk and return propertieshave been relatively well understood in the literature In Section V, we study the behaviors offund flows and the impact of fund flows on future fund performances Section VI concludes

I Data on Hedge Fund Returns and Manager Characteristics

The data on hedge fund returns and manager characteristics are obtained from TASS Amongall the datasets that have been used in the existing hedge fund literature, the TASS database isprobably the most comprehensive one TASS builds its dataset based on surveys of hedge fundmanagers Funds report to TASS mainly for marketing purposes, because they are prohibitedfrom public advertisements Overall, TASS covers more than 4,000 funds from November 1977

to September 2003 All funds are classified into “live” and “graveyard” categories “Live” fundsare those that are active as of September 2003 Once a fund is considered no longer active,

it is transferred to the “graveyard” category.5 The “graveyard” database did not exist before

1994 Thus, funds that became inactive before 1994 were not recorded by TASS To mitigatethe potential problem of survivorship bias, we include both “live” and “dead” funds and restrictour sample to the period between January 1994 and September 2003, yielding a sample of 4,131funds

Our analysis focuses on different aspects of hedge fund performances to obtain a more completepicture These include fund risk-taking behaviors (measured by overall, systematic, and idiosyn-cratic risks), raw and risk-adjusted returns, and fund flows We make these choices because

we believe that managers would devote their time and effort to improve performance measuresthat could lead to higher compensations, which could come from management/incentive fees andpersonal wealth invested in their funds For example, Goetzmann, Ingersoll, and Ross (2003)argue that both returns and capital flows are important for hedge fund manager compensation,although the relative importance depends on market condition and is time-varying The monthlyreturns provided by TASS are net of management/incentive fees and other fund expenses, and are

and Bayesian methods, the authors show that the abnormal performance of top hedge funds cannot be attributed

to luck and that hedge fund abnormal performance persists at annual horizons.

5 A fund is in “graveyard” because either it had bad performance or it had stopped reporting to TASS For instance, a fund might have done well and attracted enough capital, and it no longer has any incentive to report to TASS.

closely related to actual returns received by investors TASS also provides data on several fundcharacteristics, such as management and incentive fees, whether a fund has a high watermark,and whether its managers have personal wealth invested in the fund.

Other than returns and fund characteristics, TASS also provides rich information on personal,educational, and professional backgrounds of managers of most funds Although the return data

of TASS have been extensively studied in the literature, our paper is one of the first that examinesthe impact of manager characteristics on hedge fund performance Specifically, we identify a leadmanager of a particular fund and construct a dataset on the characteristics of this manager.6 Foreducational background, we identify the undergraduate college the manager attended, the SATscore of the college from U.S News and Princeton Review of 2003,7 whether the manager has

an MBA degree, and whether the manager has a CFA or CPA For professional background, weobtain the years the manager has worked (WORK) either directly from the dataset or assume thatthe manager started working right after MBA if he/she has one However, if neither information

is available, then WORK is missing We also obtain the number of years the manager has worked

at a particular fund, which we refer to as manager tenure (TENURE) For personal information,

we obtain the age of the manager (AGE), which is either reported in the dataset or inferred based

on the assumption that the manager was 21 upon graduation from college Generally speaking,SAT, MBA, and CPA/CFA dummies could capture either the intelligence or education of the fundmanager, while WORK, TENURE, and AGE could capture the working experience and careerconcern of the manager

Out of the 4,000 funds covered by TASS, we are able to identify most of the characteristics ofthe lead manager for 1,002 funds Panel A of Table 1 provides summary statistics on quarterlyreturns, and fund and manager characteristics for the 1,002 hedge funds.8 For fund characteristics,

we report incentive and management fees, whether the fund has a high watermark, whether themanager has personal wealth invested in the fund, the age and asset value of the fund, andthe number of managers of the fund For manager characteristics, we include SAT, MBA, andCFA/CPA dummies, AGE, WORK, and TENURE To be consistent with the Fama and MacBeth(1973) regression approach used in later analysis, we report time series averages of cross-sectionaldistributions of each individual variable That is, at each quarter, we calculate the mean, standarddeviation, minimum, first quartile, median, third quartile, and maximum of the distribution of

6 We choose the founder of a fund as the lead manager, and for funds with multiple founders we choose the one that is in charge of investment strategies or for whom the characteristics information is available.

each variable Then we report the time series averages of each of the above quantities over allquarters in our sample period.

The average raw and excess quarterly returns are 3.33% and 2.28% respectively, with a widedispersion The lowest return is around -17% and the highest is more than 26% per quarter Interms of fund characteristics, we find that most funds charge a 20% incentive fee and a 1-1.5%management fee About 40% of the funds have a high watermark, and managers of 60% of thefunds have personal wealth invested in their own funds The mean and median ages of funds areabout 4 and 3 years, respectively The mean and median fund sizes are about $86 million and

$31 million, respectively Although the majority of the funds are run by one or two managers,certain funds have more than 10 managers The SAT scores range from the lowest of 878 to thehighest of 1,511 with a mean/median around 1,300 In results not reported, about 30% of themanagers graduated from Ivy league universities About 17% of the managers have either a CFA

or CPA, and 47% of the managers have an MBA degree, while the rest fail to report on this item.9For many funds, the age variable is missing and in total we only have around 7,351 quarter-fundobservations with age information For those funds with age information, the mean and medianmanager ages are about 44 and 42.5 years, respectively, with the youngest of 27 years and theoldest of more than 72 years.10 Out of the 1,002 funds, we directly observe the WORK variablefor 899 funds For the rest of the funds, we construct WORK based on the finishing date of MBAdegree On average, managers have close to 20 years of working experience, with the shortest of

4 years and the longest of 50 years The average tenure with current fund is about 3 to 4 years,with the shortest of less than one quarter and the longest of 20 years

Panel B of Table 1 reports the correlations among fund excess returns and various fundand manager characteristics We find a positive correlation between fund excess returns andSAT, which provides preliminary evidence that managers from higher-SAT colleges are morelikely to have better performance On the other hand, we find negative correlations betweenexcess returns and fund age and several working experience variables This provides preliminaryevidence that younger funds and managers with less working experience tend to have betterperformance We find a strong positive correlation of 0.93 between fund age and manager tenure,which is consistent with the typical structure of hedge funds: They are usually established by a fewimportant managers who tend to stay with the fund.11 Chevalier and Ellison (1999b) argue that

9 A zero value of an MBA or CFA/CPA dummy variable does not necessarily mean that the manager does not have an MBA or CFA/CPA, respectively It could be that the manager fails to report this information.

10

We do not include age in our regressions because age is missing for about 40% of the funds However, due to the high correlation between age and years of working, we will not lose much information by omitting age in our analysis.

11 This result has important implications for interpreting the causality of our later finding that smarter managers tend to have higher risk-adjusted returns Although we interpret this result as evidence that smarter managers can deliver better returns, an alternative interpretation is that smarter managers are attracted to better-performing

years of working is a better proxy for working experience than manager tenure In our empiricalanalysis, we use WORK as a proxy for working experience or career concern, and we alwaysinclude fund age and lagged fund size as fund characteristics controls We also find significantpositive correlations between fund size and SAT/WORK, suggesting that manager characteristicsaffect not only the returns but also the sizes of hedge funds.

Due to the nature of currently available hedge fund datasets, most empirical studies of hedgefunds potentially face various selection biases in their data.12 To minimize the impact of sur-vivorship bias, we restrict our sample to the period between 1994 and 2003 which include bothgraveyard and live funds Panel C of Table 1 provides a comparison between graveyard and livefunds The summary statistics of graveyard and live funds are constructed in a similar way asthat in Panel A Consistent with conventional wisdom, we find that live funds tend to have higherraw/excess returns and more assets under management Although there are some differences be-tween graveyard and live funds in terms of fund and manager characteristics, these differences arenot very significant.13 Panel D of Table 1 compares the funds with manager characteristics withthe rest of the funds covered by TASS In general, we find that funds with manager characteristicstend to be younger, have higher returns and less assets under management than funds withoutmanager characteristics

II Risk-Adjustment Benchmarks for Hedge Fund ReturnsThe rich dataset constructed in the previous section allows us to examine the relation betweenhedge fund performance and manager characteristics One challenge we face in this analysis is thatrisk adjustments for hedge fund returns are much more difficult due to their use of derivativesand dynamic trading strategies Many studies have shown that standard linear asset pricingmodels fail to adequately capture the risk and return properties of most hedge funds, and it is fair

to say that there is no well-established method for hedge fund risk adjustments in the existingliterature Therefore, to ensure robust findings, we consider two broad classes of models to obtainrisk-adjusted hedge fund returns

In the first class of models, we use various hedge fund indices as benchmarks to adjust for risks

in hedge fund returns The basic idea behind this approach is that these indices might be able tocapture the risk exposures of average hedge funds and automatically adjust for the nonlinearity

hedge funds Though this interpretation could be true for mutual funds, the 0.93 correlation coefficient suggests that the hedge funds in our sample are most likely started by their current managers.

12

See Ackermann, McEnally, and Ravenscraft (1999) for a taxonomy of potential biases in hedge fund datasets.

13 One reason that live and graveyard funds have similar SAT scores is that graveyard funds include funds that have done poorly as well as funds that have done well and stopped reporting to TASS In results not reported, we divide the graveyard funds into finer sub-categories and find the liquidated funds on average have lower SATs than the graveyard funds that have done well.

in hedge fund returns One advantage of this approach is that we do not need to explicitlymodel the risk-taking behavior of hedge funds Another advantage is that this approach is easy

to implement: Investors can easily compare returns of individual hedge funds with that of broadhedge fund indices We obtain the risk-adjusted returns as the intercept term of regressions ofindividual hedge fund returns on the returns of the indices, and the risk exposures as the regressioncoefficients or the loadings of the indices

Among the three indices we consider, the first one (INDEX) is the broad hedge fund index(a weighted average of returns of all hedge funds) provided by TASS We also consider the index

of funds of funds (FoF), which is a weighted average of returns of funds of funds Fung andHsieh (2002) argue that returns of funds of funds are more accurately measured than that ofregular hedge funds and could better reflect true hedge fund performance The above two indices,however, might not be able to capture the cross-sectional differences in hedge funds strategies Forexample, TASS reports around a dozen widely followed investment styles whose risk and returnproperties differ from each other dramatically Brown and Goetzmann (2003) argue that stylescapture most of the cross-sectional differences in hedge fund returns Therefore, in addition tothe above two indices, we also use style indices (STYLE), which is the weighted average returns

of all funds within each style, to adjust the risks of hedge funds in that specific style

The second class of benchmarks we consider include the Fama-French three-factor model (FF),the model of Agarwal and Naik (2004) (AN), and the seven-factor model used in FHNR (2006).The FF model is well-established in the asset pricing literature and has been successfully applied

to returns of stocks, stock portfolios, and mutual funds The FF model has three factors: a ket factor which is the excess return of the market portfolio (MKT), a size factor which capturesreturn difference between small and big firms (SMB), and a book-to-market factor which capturesreturn difference between value and growth firms (HML) Agarwal and Naik (2004) propose toinclude option returns in traditional asset pricing models to capture the nonlinearities in hedgefund returns due to dynamic trading strategies and derivatives The AN model has two factors:

mar-a mmar-arket fmar-actor mar-as in FF, mar-and mar-an option fmar-actor which is the excess return of mar-an out-of-moneyput option on market index (OPT) We obtain the option data from CBOE Agarwal and Naik(2004) show that the AN model is relatively successful in capturing hedge fund returns Onecaveat we need to keep in mind is that option returns tend to be very volatile and could lead tonoisy parameter estimates The seven factors included in the FHNR model are the excess return

on the S&P 500 index (SNPMRF); a small minus big factor (SCMLC); the excess returns onportfolios of lookback straddle options on currencies (PTFSFX), commodities (PTFSCOM), andbonds (PTFSBD); the yield spread of the US ten-year Treasury bond over the three-month T-bill,adjusted for the duration of the ten-year bond (BD10RET); and the change in the credit spread

of the Moody’s BAA bond over the ten-year Treasury bond, adjusted for duration (BAAMTSY)

Fung and Hsieh (2004) and FHNR (2006) have shown that these factors have considerable planatory power for fund of funds and hedge fund returns.

ex-Based on the above benchmark models, we run time series regressions for each fund to estimateits risk exposures to the various factors and the risk-adjusted returns Then we take the estimatedrisk loadings and risk-adjusted returns as independent variables, and run Fama-MacBeth regres-sions on various manager characteristics More specifically, at the end of each quarter we use thepast 24 monthly returns to run the following regression:

where Ri,t is the raw return of fund i over month t, βi,q (generally a vector) represents the riskexposures of fund i at quarter q to the various factors, and Ft (also generally a vector) is themonthly value of different factors In the same regression, we also calculate the quarterly residualvolatility, ˆσi,q, as

ˆ

σi,q = [var (ˆεi,t)]1/2 with ˆεi,t = Ri,t− bαi− bβi,q0 Ft, (2)where both αbi and bβ0i,q are estimated in equation (1) In addition, we compute alpha ( ˆαi,q) andappraisal ratio (dARi,q) of fund i at quarter q, respectively, as:

Since the regression is done every quarter, we implicitly allow ˆαi,q, ˆβi,q, ˆσi,q, and dARi,q to

be time-varying This allows us to capture potential variations over time in trading strategies

of hedge funds under study While ˆβi,q measures a fund’s exposures to various systematic riskfactors, ˆσi,q measures the amount of idiosyncratic risks a fund takes While ˆαi,q measures a fund’sabnormal return, dARi,q measures abnormal return for per unit of idiosyncratic risk taken.14

To explore the relation between hedge fund performance and manager characteristics, theempirical analysis in this paper is mainly based on the Fama-MacBeth regression As an alter-native, we also conduct estimation using panel data regression with clustering and obtain similarresults Let yi,q represent one particular measure of hedge fund performance, which could beoverall return volatility, factor loadings, raw excess returns, alpha, residual volatility, appraisalratio, or fund flows of fund i at quarter q.15 Let SATi be the composite SAT score of fund i’s

14 We thank the referee for the suggestion of using residual volatility as a measure of fund performance.

15 We delete the top and bottom 1% observations on independent variables to avoid potential recording errors.

We do not conduct the bootstrap procedure of FHNR (2006) due to the small number of funds that exhibit manager characteristics.

manager’s undergraduate institute, and W ORKi,q be years of working of the manager for fund

i at quarter q, and let Controli,q be a vector of control variables for fund i at quarter q Giventhat the performance of hedge funds could depend on the size and age of the fund, we choose lagfund size and age as control variables.16 Then all empirical analysis in our paper is based on thestandard Fama-MacBeth regression approach with the following benchmark regression for eachquarter q :

yi,q= b0+ b1SATi+ b2W ORKi,q+ b03Controli,q+ ui,q (5)III Education, Career Concern, and Hedge Fund Performance

In this section, we examine the relation between hedge fund performance and manager cation and career concern, measured by SAT and WORK, respectively

edu-A Results Based on Raw Returns

Table 2 first reports the Fama-MacBeth regressions of raw excess returns on SAT and WORK

as described in equation (5) The regression results reveal a strong positive relation betweenraw excess returns and SAT The coefficient of SAT is highly significant and equals 0.091 Wealso document a strong negative relation between raw excess returns and WORK, where thecoefficient is -0.027 and highly significant The parameter estimates of SAT and WORK suggestthat everything else being the same, a manager from an undergraduate institute with a 200-point higher SAT (for instance, from George Washington University with an SAT of 1280 to YaleUniversity with an SAT of 1480) can expect to earn an additional 0.73% raw excess return peryear, and a manager with 5 years less working experience can expect to earn an additional 0.54%raw excess return more per year Given the relatively low volatility of hedge fund returns (16%per year), the difference of 0.5-0.7% in excess returns is economically important

We also examine the risk-taking behaviors of fund managers by using fund total return ity as the dependent variable in equation (5) Fund total return volatility is calculated as thevolatility of monthly returns over the past 12 months and is updated every quarter Given thatcertain hedge fund investors care about absolute performance, total return volatility is a rea-sonable measure of fund risk and has the advantage of being model free We find a significantnegative relation between fund total return volatility and SAT, suggesting that managers fromhigher-SAT institutes tend to take less risks We also find that managers with longer workingexperiences take significantly less risks

volatil-The above results are consistent with the hypothesis that better-educated managers are better

at their jobs and thus can achieve higher returns at lower risk exposures They are also consistent

16 Though it is easier to manage a smaller fund, a larger fund may have advantages in transactions costs and economy of scale Thus it is possible that there might be an optimal fund size As pointed out by Getmansky (2004), there is also a life-cycle effect in hedge fund performance.

with the career concern hypothesis that less established managers have stronger incentives towork hard at their jobs and are more willing to take risks, and consequently tend to have betterperformance than more established managers In the above regressions, we find that the controlvariable fund size is negatively related to raw excess returns and total return volatility This result

is consistent with the assumption of Berk and Green (2004) that it is more difficult to manage alarger fund given the limited number of arbitrage opportunities in the market Larger funds areusually more established and thus may have less incentives to take excessive risks Larger fundsalso can invest in more securities, which may lead to less overall volatility due to the additionaldiversification benefits

B Results Based on Risk-Adjusted Returns

Although the results in Table 2 are strong and significant, raw hedge fund returns could bedue to either risk taking or manager’s ability in identifying mispriced securities For investors whoare interested in selecting managers with positive abnormal performance, it is more interesting tostudy the relation between risk-adjusted returns and manager characteristics In this section, werelate hedge fund risk-taking behaviors and risk-adjusted returns to manager education and careerconcern To control for systematic risk, we use factor loadings and alpha as measures of risk-takingbehaviors and abnormal returns, respectively On the other hand, to control for idiosyncratic risk,

we use residual volatility and appraisal ratio as measures of risk-taking behaviors and abnormalreturns, respectively

Before we examine the cross-sectional differences in abnormal returns of hedge funds, we firstprovide some distributional statistics on alphas under different benchmark models in Panel A ofTable 3 At each quarter, we calculate the alpha of each hedge fund as in equation (3) using thesix risk-adjustment models we consider Then for each quarter, we calculate the mean, standarddeviation, and 5, 25, 50, 75, and 95 percentiles of the alphas under each model of all hedge funds.The time series averages of all the above quantities are reported for each model in the table.Under each of the six models, the average alphas are positive, and a high percentage of hedgefunds produce positive alphas Given the wide range of risk-adjustment models we consider, thisresult seems to be quite robust In addition, this result is consistent with the findings of FHNR(2006) that there are a significant number of funds of funds that produce positive risk-adjustedreturns

Panel B of Table 3 reports the results of Fama-MacBeth regressions of hedge fund alpha onSAT, WORK, fund age, and lagged fund size We find a strong positive relation between alphaand SAT, which is very robust to different risk-adjustment benchmarks we use The coefficients

of SAT in all six models range from about 0.077 to 0.174, and are mostly significant at the 5%level The one (the AN model) that is not significant at the 5% level is significant at the 10%level The parameter estimates suggest that everything else equal, a manager who graduates from

a college with a 200-point higher SAT will earn between 0.62% and 1.39% additional abnormalreturn per year We also find a negative relation between alpha and WORK, which also is veryrobust to the different risk-adjustment benchmarks The coefficients of WORK range from -0.013

to -0.027, and most of them are highly significant The only exception is that of the AN model,

in which the large standard errors could be driven by the volatile option factor The parameterestimates suggest that a manager with 5 years less working experience can earn between 0.26%and 0.54% additional abnormal return per year

Panel C of Table 3 contains the results of Fama-MacBeth regressions of estimated fund riskloadings under different models on SAT, WORK, fund age, and lagged fund size Though thedependence of factor loadings on SAT and WORK is not as uniform as that of alpha, we do find astrong negative impact of SAT on the risk-taking behaviors of hedge funds For example, a higherSAT leads to significantly lower factor loadings of eleven out of the fifteen risk factors included inthe six benchmark models Although the results for WORK are not as strong and uniform as thatfor SAT, they still point to a general negative relation between factor loadings and WORK Ahigher WORK leads to significantly smaller factor loadings for eight out of the fifteen risk factors.The results for most other factors are negative, but not statistically significant

Panels D and E of Table 3 contain the results of Fama-MacBeth regressions of residual ity and appraisal ratio under different models on SAT, WORK, fund age, and lagged fund size,respectively We find a strong negative relation between residual volatility and both SAT andWORK, although the impact of WORK is much smaller than that of SAT We also find a strongpositive relation between appraisal ratio and SAT On the other hand, we find a negative relationbetween appraisal ratio and WORK Therefore, better-educated managers not only take less idio-syncratic risks, they also earn higher abnormal returns for per unit of idiosyncratic risk taken Incontrast, although more established managers take less idiosyncratic risk, they earn less abnormalreturns for per unit of idiosyncratic risk taken as well

volatil-The SAT score of a manager’s undergraduate institute could represent different qualities of themanager For example, higher SAT could mean the manager is more intelligent, more ambitiousand driven, more competitive, has better work ethics, or better educated, etc.17 Given the highlycompetitive nature of the hedge fund industry and the complexity of hedge fund strategies, ourresults on hedge fund performance and SAT are consistent with the conjecture that managerswith better educational backgrounds might be able to understand, design, and implement thesestrategies better than others, either because they are smarter and better educated, or becausethey are more devoted to their jobs

Manager working experience could measure a manager’s knowledge and experience about

17 SAT also could measure how closely connected graduates from a certain university are We use endowments per student for each university as a proxy for connection and find that it has no significant impact on performance.

the industry, as well as the manager’s incentive to work hard at his/her job On one hand, amore experienced manager might be able to earn higher returns due to his/her experience andknowledge On the other hand, because such a manager is more likely to be better established,he/she also may have less incentive to work hard than a manager who still needs to establishhis/her career The negative relation between hedge fund performance and WORK seems tosuggest that the impact of career concern dominates that of working experience.

Collectively, the results in Table 3 suggest that better-educated and more established managerstend to take less systematic and idiosyncratic risks than their peers Moreover, better-educated(more established) managers also earn higher (lower) abnormal returns for per unit of systematicand idiosyncratic risks taken These patterns strongly suggest that certain managers are indeedbetter than others and, no matter seeking superior absolute or relative performance, investors arebetter off by selecting less established managers with better educational backgrounds, everythingelse the same FHNR (2006) also document significant cross-sectional differences in risk-adjustedreturns of funds of funds Our results extend FHNR (2006) by relating differences in hedge fundperformances to education and career concern, and therefore provide a guidance on identifyingsuperior hedge fund managers based on manager characteristics

First, we repeat all the analysis in the previous section using data on funds of funds According

to TASS, about 10% of hedge funds belong to the so-called funds of funds Though regular hedgefunds make direct investments in different markets, funds of funds invest in a group of other hedgefunds By doing this, funds of funds can diversify away idiosyncratic risks in individual hedgefunds and thus achieve more stable returns The incentive fees of funds of funds (typically around10%) are also significantly lower than that of regular funds (typically 15%-20%) One importantadvantage of funds of funds, as pointed out by Fung and Hsieh (2004) and FHNR (2006), isthat their returns are less susceptible to survivorship bias than regular hedge fund returns mighthave.18Although our previous analysis includes both live and dead funds, results based on funds18

Fung and Hsieh (2000) and FHNR (2006) argue that returns of funds of funds are less susceptible to survivorship bias issues than that of regular hedge funds Suppose the returns of a regular hedge fund that goes out of business are not recorded in a dataset Then analysis based on the returns of regular hedge funds contained in the dataset would suffer from survivorship bias On the other hand, the returns of a fund of funds that has invested in the defaulted hedge fund would partially reflect the negative returns of the defaulted fund Because of the more diversified investments of funds of funds, they are more likely to survive than regular funds, and survirorship bias

of funds further ensure that our results are not driven by survivorship bias.

The results for the 122 funds of funds in our sample in Panel A of Table 4 are generallyconsistent with that of the regular hedge funds in Tables 2 and 3.19 For example, we find asignificant positive (negative) relation between excess returns (residual volatility) and SAT Thenegative-but-not-significant relation between total volatility and SAT could be due to the factthat funds of funds are generally pretty well diversified We also find a positive relation betweenboth alpha and appraisal ratio and SAT Although the coefficients of SAT in both regressions arenot significant at the 5% confidence level, they are significant at the 10% level We find a muchweaker negative impact of WORK on alpha, residual volatility, and appraisal ratio for funds offunds than for regular funds: The Fama-MacBeth coefficients of WORK are not significant in allthree regressions The potential financial rewards from the high incentive fees of regular fundsmight provide stronger incentives for less established managers to work harder in the early stage

of their careers This could lead to a bigger decline in performance when these managers start toslow down after they become more established On the other hand, because such incentives forless established managers are weaker for funds of funds, the declines in performance as managersbecome more established also could be less dramatic

Second, we repeat all the analysis using value-weighted average returns of all hedge fundsmanaged by the same manager Sometimes the same manager may start more than one fundwith different fund structures.20 One of these funds may be used as a “showcase” fund, which haslow assets under management but great performance Unfortunately, these funds are closed andare used as a marketing tool to attract capital in other less well-performing funds launched by thesame manager Our previous analysis would treat these funds as multiple observations, although

in reality they are managed by the same manager To control for this effect, we study the impact

of manager characteristic on the weighted average returns of all funds managed by the samemanager The results of this analysis, reported in Panel B of Table 4, are broadly consistent withthat of our original analysis We find a strong positive impact of SAT on excess return, alpha, andappraisal ratio Given that a portfolio of hedge funds is generally better diversified than a singlefund, the impact of SAT on volatility is weaker for individual managers For example, although

we find a negative relation between SAT and residual volatility, the coefficient is significant only

at the 10% but not at the 5% level Moreover, we do not find any significant relation betweenoverall volatility and SAT We also find weaker impact of WORK on various aspects of hedgefund performance for individual managers than regular hedge funds

is a lesser concern for funds of funds.

Finally, we repeat all the analysis for two different subperiods during our sample to controlfor time-varying risk exposures in hedge fund returns Due to the dynamic nature of hedge fundbusiness, it is highly possible that hedge fund strategies, risk exposures, and risk-adjusted returnsall change over time Moreover, recent studies of Fung and Hsieh (2004) and FHNR (2006) showthat there are structural breaks in hedge fund returns caused by LTCM crisis (around October1998) and bursting of the Internet bubble (April 2000) Although our Fama-MacBeth regressionsexplicitly allow for time-varying risk-adjusted returns and risk exposures, sub-period analysisprovides further assurance that our results are robust to these structural breaks Specifically, werepeat our analysis for the following two subperiods in which hedge fund returns are relativelystable: Q1.1995 to Q2.1998 and Q3.2000 to Q3.2003 We ignore the period between Q3.1998 andQ1.2000, because the few quarters of observations during this period makes it difficult to conductFama-MacBeth regressions For brevity, we only report results based on the FHNR model inPanel C of Table 4 We find the results in the two subperiods are generally consistent with eachother For example, we find a significant negative (positive) relation between residual volatility(appraisal ratio) and SAT in both sub-periods Although we find a significant positive relationbetween alpha and SAT in the first sub-period, the coefficient of SAT in the second sub-period

is significant only at the 10% level Furthermore, the SAT coefficients on alpha and appraisalratio in the second sub-period are much smaller than that in the first sub-period This result

is consistent with the finding of FHNR (2006) that most hedge funds perform quite poorly andexhibit small cross-sectional differences in their performances during the second sub-period.21

IV Two Special CasesAlthough the relations between hedge fund performance and SAT/WORK are pretty robust tovarious risk-adjustment benchmarks, these models could still be misspecified In this section, wespecialize our analysis to two specific hedge fund strategies whose risk-return characteristics havebeen reasonably well understood in the literature These two strategies are the trend followingstrategy studied by Fung and Hsieh (2001) and the risk arbitrage strategy studied by Mitchelland Pulvino (2001)

A Trend Followers

The trend following strategy has been widely used by commodity trading advisors, who hope

to identify trends in prices and then either buy on upward or sell on downward trend Thereturns on trend followers tend to be large and positive during the best and worst periods ofmarket performance Due to this nonlinear feature, the returns on trend followers have lowcorrelations with standard equity, bond, currency, and commodity indices The important work

of Fung and Hsieh (2001) shows that returns of trend followers resemble closely that of a lookback

21 We thank the referee for recommending this explanation of the result to us.

straddle They further demonstrate that the following model, which includes lookback straddlesfrom several major markets (stock, bond, commodity, interest rate, and foreign currency), canexplain the returns of trend followers very well:

Ri,t = αi+ βi,stockP T F Sstock,t+ βi,bondP T F Sbond,t+ βi,currencyP T F Scurrency,t

+βi,interestP T F Sinterest,t+ βi,com mod ityP T F Scom mod ity,t+ ei,t, (6)where PTFS stands for lookback straddle returns, and the subscripts represent the market forwhich the straddles are constructed

We apply the model of Fung and Hsieh (2001) to obtain the risk-adjusted returns of the 90live and dead funds in the TASS dataset that have a “trend following” investment focus and forwhich we can identify the manager characteristics In addition to the model of Fung and Hsieh(2001), we also construct a trend following index based on the value weighted average returns ofthe 90 trend following funds This index allows us to measure whether an individual fund takesmore or less style risk than the average trend following fund

In Panel A of Table 5, we estimate three different versions of the model for the average returns

of the 90 trend following funds In the first model, which includes the lookback straddles from allfive markets, the adjusted R2is 27.8% and the coefficients of currency and commodity straddles arehighly significant The next two models use only a subset of the five straddle factors Consistentwith Fung and Hsieh (2001), we find that the incremental contributions of stock and interest ratestraddles over the other three are very small We obtain similar results in later analysis using allthree models and report results based on the second model which has the highest adjusted R2.Panel B of Table 5 reports the results of Fama-MacBeth regressions of excess returns, factorloadings, alpha, residual volatility, and appraisal ratio based on the Fung and Hsieh (2001) model

on SAT and WORK The relation between excess returns and manager characteristics for trendfollowers becomes somewhat weaker For example, we do not find significant dependence ofexcess returns on either SAT or WORK However, consistent with previous results, we find astrong positive effect of SAT on both alpha and appraisal ratio For example, the SAT coefficient

in alpha regression, which is 0.365, suggests that a manager from an undergraduate institute with

a 200-point higher SAT can earn a higher alpha of almost 2.9% per year On the other hand, we

do not find a strong relation between alpha and WORK for trend followers

The results in Panel C of Table 5 based on the trend following style index are generallyconsistent with that in Table 3 We find managers from higher-SAT institutes take significantlyless style risk than the average trend following fund These managers also earn significantly higheralpha and appraisal ratio Although we find a positive relation between residual volatility andSAT in both Panels B and C, the SAT coefficients are not significant in either case

Although the model of Fung and Hsieh (2001) serves a good risk-adjustment benchmark for

trend followers, there are some caveats we need to keep in mind when interpreting the results.First, due to the relative small number of trend following funds in our sample, we may not haveenough information to obtain statistically significant estimates Second, the returns on straddlestend to be very volatile, which also could introduce noises in parameter estimates Indeed wefind a weaker relation between hedge fund performance and manager characteristics for trendfollowers Despite that, we still find manager educational background to be positively related torisk-adjusted returns Therefore, the overall results of Table 5 suggest that the relation betweenhedge fund performance and manager characteristics documented for regular hedge funds generallyhold for trend following funds.

B Risk Arbitragers

Risk arbitrage, or the so-called merger arbitrage, has become very popular among hedge funds.Generally speaking, in a merger or takeover the acquiring company’s stock price tends to go downand the acquired company’s stock price tends to go up To take advantage of this pattern, riskarbitragers typically short the acquiring company and long the acquired company Although thestrategy is called merger arbitrage, it is not a pure arbitrage opportunity The main risk is thatthe deal may not go through, in which case the arbitragers would lose money

Mitchell and Pulvino (2001) have conducted a careful analysis on the risk and return properties

of merger arbitrage They examine a return index of all the merger deals in the past few decadesand returns of hedge funds that specialize in this strategy They find that returns on mergerarbitrage resemble that of a shorted put option That is, the strategy makes money if the marketgoes up or at least stays flat, but loses money if the market goes down Mitchell and Pulvino(2001) use the following piecewise linear regression to capture the returns of the risk arbitragestrategy,

Ri,t = (1 − δ)(αi,low+ βi,lowM KTt)I(M KTt< −4%)

+δ(αi,high+ βi,highM KTt)I(M KTt > −4%) + εi,t (7)where Ri,t is the excess return on asset i at time t, δ is a weighting coefficient, αi,low (αi,high) isthe intercept when market return is below (above) -4%, βi,low (βi,high) is the market risk loadingwhen market return is below (above) -4%, M KTtis the excess return of the market portfolio, andI(·) is an indicator function, which equals one if the condition in the parenthesis is true and zerootherwise Mitchell and Pulvino (2001) show that by allowing different exposures to the marketfactor in up and down markets, the above model captures the returns of merger arbitrage verywell

We apply the model of Mitchell and Pulvino (2001) to the 150 funds in the TASS dataset thathave an investment focus of “risk arbitrage” and for which we can identify most of the managercharacteristics Although the sample period of Mitchell and Pulvino (2001) is between 1963 and