journal of investment management joim · a risk management question, ... after showing how...

JOIMwww.joim.com

JOURNAL OF INVESTMENT MANAGEMENT, Vol. 8, No. 1, (2010), pp. 1–25

© JOIM 2010

DO ENDOWMENT FUNDS SELECT THE OPTIMAL MIXOF ACTIVE AND PASSIVE RISKS?

Keith C. Brown ∗,a and Cristian Tiub

The investment decision confronting managers of multi-asset class portfolios can be characterizedin terms of the passive (i.e., benchmark or policy) and active (i.e., market timing and securityselection) strategies they adopt. In this paper, we investigate whether managers select the appropriatecombination of active and passive allocations in their portfolios. Noting that this issue is ultimatelya risk management question, we adapt a simple framework for establishing what constitutes theoptimal level of active and passive risk exposures. We then examine the question empirically usinga database consisting of the allocation decisions and investment performance of a large set ofuniversity endowment funds over the period from 1989 to 2005. Our findings show that (i)the average endowment had too little active risk exposure in its portfolio, (ii) endowment fundscould have significantly increased their risk-adjusted performance by enhancing the scale of thealpha-generating strategies they were already employing, and (iii) this tendency to underutilizeactive management skills was more pronounced for larger endowments than for smaller ones. Weconclude that the typical endowment fund could have improved its performance by increasing thecommitment to its active management skills.

1 Introduction

It is well established that the performance of aninvestment portfolio that is diversified across mul-tiple asset classes can be thought of as beingdriven by three distinct decisions that its manager

∗Corresponding author.aMcCombs School of Business, B6600, The Universityof Texas at Austin, Austin, TX, 78712, USA. Email:[email protected] Management Center, University at Buffalo, Buffalo,NY 14260, USA. Email: [email protected]

makes: (i) long-term (i.e., strategic or policy) assetallocations; (ii) temporary adjustments (i.e., tac-tical) to these strategic allocations in response tocurrent market conditions; and (iii) the choice of aparticular set of holdings to implement the invest-ment in each asset class (i.e., security selection). Thefirst of these performance components is commonlyreferred to as the passive portion of the portfoliowhile the latter two collectively represent the activepositions the manager adopts. Trying to under-stand and disentangle the issues defining the myriaddimensions of the relationship that exist between

FIRST QUARTER 2010 1

2 KEITH C. BROWN AND CRISTIAN TIU

these passive and active performance componentshas become, perhaps, the preeminent challengeconfronting the portfolio management industry.

In this study, we extend this “active–passive” discus-sion by considering the following question: Giventhat both active and passive risks play a role in deter-mining the overall performance of a portfolio, isit possible to establish what the optimal combina-tion of these components should be? To addressthis problem, we create a simple analytical frame-work that expands the pioneering work of Treynorand Black (1973) (hereafter T–B) to a multi-assetclass setting. Specifically, after showing how port-folio returns can be decomposed along the linesof the three managerial decisions outlined above,we adopt the central T–B notion that the activeand passive portions of any collection of securi-ties can themselves be viewed as separate assets thatcan be combined in an optimal manner. Assum-ing an appropriate risk-adjusted decision rule—thereward-to-variability and reward-to-downside riskratios, in our case—then allows us to calculate theoptimal passive and active allocations in the port-folio by finding the levels of those components thatmaximize the decision metric.

To demonstrate the insights gained by this opti-mization process, we analyze the investment per-formance of a sample of college and universityendowments funds over the period from 1989 to2005. Several salient characteristics of the endow-ment fund environment make these investmentvehicles particularly well-suited to this purpose,including the fact that they have virtually infinitetime horizons, they tend to invest across as many as adozen different asset classes, and they have shown ahistorical commitment to pursuing both passive andactive strategies (see, for instance, Swensen, 2009).Making use of this database allows us to pose twospecific queries. First, based on an in-sample anal-ysis of the entire database, did the level of activerisk that the average endowment fund manager

assumed constitute an optimal level of exposure?Second, using an out-of-sample analysis that splitsthe database into half chronologically, could thetypical endowment manager have improved per-formance in the second half of the period byaltering the portfolio’s active–passive risk combi-nation based on information revealed in the firsthalf?

Our empirical analysis strongly suggests that theanswers to these questions are “no” and “yes,”respectively. That is, over the entire sample period,we document that the average endowment fund wasoverallocated to the passive component in the port-folio by as much as 30 percent, meaning that itseverely underemphasized the active managementskills that were already in place. Further, based onour out-of-sample analysis, we demonstrate that ifthe typical endowment manager had adjusted theportfolio’s passive and active risk allocations basedon the findings from the first half of the sampleperiod, the risk-adjusted performance during thesecond subperiod would have improved by about20 percent. We stress, however, that these real-locations from passive to active risk exposure donot necessarily require the average endowment toincrease the net level of its transactions or take on asignificantly greater level of total risk—particularlyif the active and passive components are negativelycorrelated—nor does it suggest an increased propor-tional exposure to asset classes that are perceivedto be riskier (or otherwise objectionable), such ashedge funds or private equity.

To assess the robustness of these findings, wereplicated the analysis on various stratificationsof our sample based on cross-sectional differencesin characteristics that have been shown to mat-ter for endowment funds. Specifically, followingBrown et al. (2009), we divide the fund sampleinto extreme values of assets under managementand spending ratio, as well as distinguish thosefunds associated with public versus private academic

JOURNAL OF INVESTMENT MANAGEMENT FIRST QUARTER 2010

DO ENDOWMENT FUNDS SELECT THE OPTIMAL MIX OF ACTIVE AND PASSIVE RISKS? 3

institutions. The results indicate that while someof these characteristics appear to significantly influ-ence the active–passive allocation decision, othersdo not. We show that the size of the fund makesan appreciable difference, with the optimal activerisk allocation for the largest funds being almostdouble what they actually hold whereas small fundsare shown to be exploiting their active manage-ment talents more completely. On the other hand,cross-sectional variations in endowment spendingrates or the profile of the sponsoring universityhad no material impact on the findings for theoverall sample. Overall, we conclude that the typi-cal endowment fund—and particularly the typicallarge fund—could have improved its performanceby taking more active risk while keeping its policyallocation unaltered.

The remainder of this paper is organized as follows.In the next section, we review the relevant literatureas well as develop our simple framework for assess-ing the optimality of a manager’s active–passiveallocation decision. In Section 3, we describe ourendowment fund database while Section 4 presentsour main empirical findings on both a full-sampleand cross-sectional basis. Section 5 then addressesseveral normative implications concerning the abil-ity of investors to implement this framework inpractice while Section 6 concludes the study.

2 Finding the optimal mix of activeand passive risks

2.1 A brief overview of the risk budgeting process

One of the most important recent trends in themanagement of investment portfolios has been thedevelopment of formal procedures to help man-agers determine the appropriate combination of riskexposures that they should be holding. In particular,in what has come to be referred to as the risk bud-geting process, considerable attention has been paidto establishing the conceptually correct mixture of

passive (i.e., systematic) risk and active (i.e., unsys-tematic or residual) risk that defines the portfolio.Indeed, as characterized by Chow and Kritzman(2001), an investor’s risk budget can ultimately beviewed as a means of converting her asset alloca-tion decisions—expressed in monetary terms—intothese risk assignments.1

Although they differ in their specific details, a com-mon feature of the various extant risk budgetingmethodologies is that the optimal combination ofactive and passive risks is established as the one thatmaximizes the total risk-adjusted expected returnsfor the portfolio. In this context, Clarke et al.(2002) make the crucial observation that while thetotal investment return is simply the sum of thereturns generated by the active and passive partsof the portfolio, the portfolio’s total risk is notadditive but rather depends on: (i) the volatil-ity of passive returns, (ii) the volatility of activereturns, and (iii) the correlation that exists betweenthe passive and active return components. Whilesome risk allocation models ignore this interactionterm (e.g., Waring and Siegel, 2003 cite historicaldata to support their assumption that these correla-tions are insignificantly different from zero), otherstake it into account. For instance, Berkelaar et al.(2006) allow for non-zero active risk interactionsin the development of their correlation-adjustedmethod for maximizing a portfolio’s risk-adjustedperformance.

A notable feature that appears to be pervasivethroughout the risk budgeting literature is that theredoes indeed seem to be a role for taking active riskin the portfolio. This is not a trivial outcome giventhat, in equilibrium, the overall contribution fromactive management should be a zero-sum game.2

However, the fact that the net expected return tothe active risk-taking is zero does not mean that allactive managers will fail to produce a positive alpha(i.e., the difference between the realized and theexpected return). Thus, as Winkelmann (2003) and

FIRST QUARTER 2010 JOURNAL OF INVESTMENT MANAGEMENT


Leibowitz (2005) note, the case for active portfoliorisk rests on markets not always being in equilibriumand the investor’s ability to identify genuinely skill-ful active managers in advance. With respect to thelatter condition, Grinold (1989) and Grinold andKahn (2000) have established a useful frameworkfor assessing the expected risk-adjusted contribu-tion of active managers through the skill they areperceived to possess and the breadth with whichthey can deploy those skills.3

Of course, such justifications for including activerisk elements in an investment portfolio are not new.Often lost in the risk budgeting discussion is the factthat Treynor and Black (1973) addressed this topicalmost four decades ago. Nominally an analysis ofthe role that security analysis can play in marketsthat are almost efficient, the T–B framework pro-vides a compelling way to think about how passiveand active allocations should be pooled. The essen-tial insight into the T–B analysis is that the optimalcombination of the active portfolio—which resultsfrom the application of security analysis to iden-tify a limited number of undervalued assets—anda passive benchmark portfolio is itself a straightfor-ward portfolio optimization problem. That is, T–Btreats the active and passive portions of an invest-ment portfolio as two separate “assets” and thencalculates the mix of those assets that maximizesthe reward-to-variability (i.e., Sharpe) ratio. It isthen demonstrated that the investment allocationassigned to the active portfolio strategy increaseswith the level of alpha it is expected to produce (i.e.,the active “benefit”), but decreases with the degreeof unsystematic risk it imposes on the investmentprocess (i.e., the active “cost”). This is an importantinsight because it suggests that taking more activerisk in a portfolio will not necessarily lead to anincrease in total risk; if, for instance, the manager’sactive investment is negatively correlated with thepassive component (e.g., an effective short positionin an industry or sector benchmark) the overall riskin the portfolio might actually decline.

For all of its elegance, Kane et al. (2003) have notedthat the T–B model has had a surprisingly low levelof impact on the finance profession in the yearssince its publication. They attribute this neglect tothe difficulty that investors have in forecasting activemanager alphas with sufficient precision to use theT–B methodology as a means of establishing mean-ingful active and passive portfolio weights on anex ante basis. However, this begs the question ofwhether the model offers a useful way of assessingex post whether the proper active–passive allocationstrategy was adopted by the portfolio manager. Saiddifferently, using the T–B model, did the manageraffect an appropriate combination of active and pas-sive exposures, given the alphas that were actuallyproduced? In the subsequent sections, we providejust such an empirical analysis using our sample ofendowment funds.

2.2 Decomposing portfolio returns

In order to address the issue of whether fund man-agers deploy the various risks in their portfolio inan optimal manner, it is first necessary to split thereturns they produce into their passive and activecomponents. We follow a methodology similar toBrinson et al. (1986) and Brinson et al. (1991) anddecompose the returns of a managed portfolio intotheir three fundamental components: (i) strategicasset allocation policy (i.e., benchmark), (ii) tacticalallocation (i.e., market timing), and (iii) secu-rity selection. As noted earlier, this decompositionreflects closely the investment choices within a typ-ical endowment fund; the first of these componentsrepresents a passive decision typically made by theendowment’s Board while the latter two are activedecisions made by the endowment’s investmentstaff.4

Formally, let Ri,t be the total realized return on fundi at the end of period t , wi,j,t the actual portfo-lio weight of fund i in asset class j at the end ofperiod t and ri,j,t the period-t return on asset class j.



The realized return can then be expressed as:

Ri,t =N∑

j=1

wi,j,t−1 ri,j,t , (1)

where N denotes the set of investable asset classes.

The policy allocation return of a fund is the partof the total return that can be achieved by passivelyapplying the strategic policy asset allocation weightsto the benchmark indices for each asset class. Inother words, no skill in timing exposures to thevarious asset classes or selecting securities withineach class is required to achieve this return. Thebenchmark return for fund i at the end of period tis formally defined as follows:

RBi,t =

N∑j=1

wBi,j,t−1rB

j,t , (2)

where wBi,j,t−1 denotes the asset allocation policy

weight in asset class j at the end of period t − 1and rB

j,t is the period-t buy-and-hold return on thebenchmark index for asset class j. Note that deter-mining the level of Eq. (2) requires only knowledgeof (i) the policy asset allocation weights and (ii) thebenchmark index for each asset class.

We define the market timing component, the firstof the two active return elements, as follows:

RTi,t =

N∑j=1

(wi,j,t−1 − wBi,j,t−1)rB

j,t , (3)

where, as before, wi,j,t−1 denotes the actual weightin class j at the end of period t − 1 and wB

i,j,t−1 isthe corresponding policy weight for the same invest-ment horizon, decided one period in advance. Asformulated, this expression captures the return thatis achieved by over- or underweighting the bench-mark asset in an attempt to increase returns orreduce risk.

The last component of a portfolio’s total return cap-tures the effect of security selection, which is the result

of the second active decision the portfolio managercan make. This is the excess return of the managedportfolio in a given asset class over the hypothet-ical return achievable by an investor who allocatesresources in the benchmarks according to the policyweights. Formally, we define the security selectionreturn as:

RSi,t =

N∑j=1

wi,j,t−1(ri,j,t − rBj,t ), (4)

where ri,j,t is the return generated by a portfolioof securities chosen by the manager responsible forasset class j.

From the decompositions in Eqs. (2)–(4), it is clearthat the portfolio’s total return in any period t canbe expressed as the sum of its passive and activecomponents or:

Ri,t = RBi,t + RT

i,t + RSi,t , (5)

which, in turn, implies that the total active portionof the portfolio’s return can be obtained as the dif-ference between its overall return and the passivecomponent:

Ri,t − RBi,t = RT

i,t + RSi,t = RA

i,t . (6)

Assuming that RBi,t adequately expresses the portfo-

lio’s expected return during period t , then RAi,t in

Eq. (6) represents the manager’s periodic realizedalpha from his or her active decisions.5

2.3 Optimal active–passive allocation: A simpleframework

Following the intuition of Treynor and Black(1973), we address the question of whether port-folio managers adopt an optimal active–passive riskallocation by seeing if they take full advantage oftheir alpha-generating capabilities or whether they“leave money on the table” by mixing their passiveand active positions in a sub-optimal manner. Inwhat follows below, we present a straightforward



extension of the T–B model that allows us to assessthis issue in the context of the multi-asset class prob-lem faced by endowment fund managers who havethe ability to make active decisions about broadmarket and sector exposures as well as for individualsecurity positions.

To formalize the evaluation process, assume asbefore that the total actual return R of a particularendowment contains a passive benchmark com-ponent RB as defined in Eq. (2) and an activecomponent RA representing, without loss of gen-erality, the collective impact of the timing andselection decisions the manager makes. In this sim-plification, the active component can be written asRA = R − RB, where the subscript denoting theinvestment period is suppressed for convenience.Notice that a fund can achieve exposure to its bench-mark either by investing directly in the indicescomposing RB or indirectly through the formationof the active portfolio that generates R. Conse-quently, we can always think of this active portfolioitself as being a (trivial) combination obtained byinvesting 100% in the assets that deliver R and 0%in the assets that delivers RB.

It is important to realize, though, that this “allactive” portfolio will have only indirect exposureto the benchmark through R. The crucial insightinto this setting is that by rescaling the existingpositions in R and RB, the portfolio manager canconstruct an alternative portfolio that has the samemonetary commitment to the benchmark assets butachieved with a different combination of asset classand security exposures. For example, consider thenew return:

R(λ) = λR + (1 − λ)RB. (7)

The portfolio implied by the weighted return inEq. (7) is the de facto result of a swap transac-tion between the existing active portfolio and thebenchmark allocation. If, for instance, the actualinvestment weights in the original portfolio and inthe benchmark are identical (i.e., in the absence

of market timing), the resulting swap portfolio hasthe same asset allocation weights, but its exposure tothe benchmark is achieved through different secu-rities than those contained in the active portfoliodelivering R.

To implement such an exchange, an endowmentdoes not have to invest in asset classes with whichit is not already familiar. To see this, consider thecase of a portfolio comprising a single asset class,US domestic equities. Suppose that the managerinitially holds $100 million of US equity in Port-folio R and then decides to revise his position bychoosing an allocation of 110% in R(λ = 1.1),while simultaneously shorting 10% of the bench-mark. After this swap, the new portfolio will contain$110 million of this new equity position and willbe short $10 million of the benchmark for USequity (e.g., the CRSP value-weighted portfolio, theWilshire 5000 or the Russell 3000). The net overallUS equity position is still $100 million and so theexposure to this asset class is unaltered, althoughthe actual securities held in the adjusted managedportfolio will be different. For the swap to be imple-mentable in practice, it is important that it doesnot alter substantially the overall exposure to anasset class, since most institutional investors havepolicies limiting the variation of their actual port-folio weights around their benchmark weights (i.e.,tactical ranges). Furthermore, the implementationof this swap does not necessarily require short sell-ing the benchmark index; and it merely requiresthe sale of 10% of a combination of securities ina portfolio that is close to the index, while simul-taneously investing proportionally the proceedingsfrom this sale in those securities remaining in theportfolio. Hence, no new active management skillsare required beyond those the manager alreadypossesses.

Since by definition R = RB + RA, using Eq. (7)the passive component of the post-swap portfo-lio is RB(λ) = RB while the active component is



RA(λ) = λRA. By extension, then, when λ > 1 theswap portfolio will have a higher emphasis on theactive management component than did the man-ager’s original portfolio. In other words, choosinghow much emphasis is best placed on active risk isequivalent to choosing how to best rescale the exist-ing managed portfolio R by swapping a fraction of itagainst the benchmark portfolio RB. It is importantto note that although the implementation of such astrategy requires the actual portfolio to be scalable,it does not require any additional alpha-generatingabilities compared to the actual portfolio R.

With this background, the specific question wewould like to consider can be stated as follows: Forany particular endowment fund, is it possible toproduce a portfolio with a more pronounced com-mitment to active risk that would be preferred tothe portfolio currently held? If so, we can concludethat such an endowment is passing on a readilyavailable opportunity to use its active managementskills in a way that could potentially add valuethrough the market timing or security selectionreturn components.

To answer any such question concerning the opti-mality of the investment process, we first need toidentify endowment preferences and then solve forthe rescaling parameter λ that maximizes those pref-erences. We specify two different characterizationsof a fund’s objective function, both of which assumethat the investor is best served by the portfolio posi-tion that maximizes risk-adjusted returns. In thefirst specification, we follow the T–B approach ofassuming that endowments select an active and pas-sive mix in order to maximize the reward-to-totalvariability (i.e., Sharpe) ratio of the portfolio, or:

λ = arg maxE [R(λ) − Rf ]std[R(λ) − Rf ] , (8)

where E [·] and std[·] are estimated by their samplecounterparts.6

Of course, one challenge with using the objectivefunction in Eq. (8) is that it is based on the implicitassumption that the return standard deviation isan adequate characterization of risk. However, sev-eral of the asset classes frequently included inendowment fund portfolios are likely to producereturns that are asymmetrically distributed (e.g.,hedge funds).7 Accordingly, the second optimiza-tion problem we examine assumes that endowmentsselect active and passive components by maximiz-ing the reward-to-downside risk (i.e., Sortino) ratio,defined as:

λ = arg maxE [R(λ) − Rf ]

DR(λ), (9)

where

DR(λ)

=√

E [max{E [R(λ) − Rf ] − (R(λ) − Rf ), 0}2]is the downside risk measure, also computed via itssample counterpart. This second criterion can alsobe justified as being relevant to a university endow-ment that is concerned with producing returnsin the portfolio that meet or exceed its spend-ing requirement. In this case, one can argue thatthe semi-deviation, or downside risk, is the morerelevant measure of uncertainty.

3 Data description

Our primary database is the set of endowment stud-ies produced by the National Association of Col-lege and University Business Officers (NACUBO),which are annual publications based on surveys thatgather information about net-of-expense returnperformance, asset allocation patterns, spendingrules and rates, and manager and custodial rela-tionships of college and university endowmentsthroughout the United States, Canada, and PuertoRico. The basic set of data available to us from thissource covers the period from 1984 to 2005.8

Although the NACUBO surveys began in 1984,the participating institutions were not identified



during the 1984–1988 period meaning that, forthese five years, the asset allocation survey data can-not be merged with the information on assets undermanagement, endowment fund payout or invest-ment performance. This in turn means that the twocomponents of active returns described above can-not be measured accurately and, as a consequence,the majority of our analysis will be limited to thepost-1988 period. For the 1989–2005 time frame,identification of member endowments is possibleand we obtained this information from NACUBOdirectly. Although the NACUBO studies are pub-licly available, identification of the members isnot.

An important adjustment in the NACUBO sur-veying process during our sample period involvesthe collection of information on both the actual aswell as the intended (i.e., policy) asset allocationschemes. In their surveys during the 2002–2005period, NACUBO asked participating endow-ments to report not only their actual asset allocationbut also their target levels for the next year. In thework to follow, we interpret this target allocation asderiving from the fund’s policy, inasmuch as it repre-sents the desired exposure to the various asset classesthat is usually decided by the board of the institutionas a general mandate for the investment process. Foryears prior to 2002, we take as proxies for the targetweights the actual portfolio weights for each fundlagged by one year (i.e., wB

i,j,t−1 = wi,j,t−2).

Table 1 provides an overview of some of the morenotable characteristics of our endowment fund sam-ple. Panel A of the display lists summary statisticsfor the full sample while Panels B and C give sep-arate summaries for the endowments of privateand public institutions, respectively. As shown inPanel A, the endowment fund universe increasedsignificantly over the sample period; the number offunds more than tripled (from 200 to 709) and therewas a roughly tenfold increase in the aggregate assetsmanaged by the industry (from $25.4 billion to

$249.3 billion). Second, the assets under manage-ment (AUM) for the largest and smallest funds (e.g.,$25.5 billion versus $1.26 million in 2005) indi-cates the tremendous cross-sectional heterogeneityin the sample and suggests that endowments ofdifferent sizes may face very different asset man-agement problems that require different mixturesof active and passive risks.

The data in the first panel of Table 1 also show thatwhile the overall distribution of fund returns arenot highly skewed (i.e., there is not a large dis-crepancy between the mean and median returnsreported for most years), the differences betweenthe best and worst performing funds are typicallyconsiderable. Further, the last four columns of thedisplay also show that there is a substantial degreeof variation across endowments in terms of theirspending needs. The average annual payout, whichNACUBO only began reporting in 1994, is about5.0%, but the spread of values in a given year rangesfrom virtually zero to almost 25%. This is impor-tant in the present context because fund spendingpolicies may influence the manager’s risk allocationdecision for the portfolio.

Another way in which endowment funds appearto differ in a meaningful way is whether the spon-soring educational institution is private or public.Panels B and C of Table 1 show these data startingin 1989, which is when NACUBO began recog-nizing this distinction. For any given year, thereare three-to-four times more private school endow-ments than public ones and the average AUM forthe former group is always larger. Additionally, pri-vate school funds generated a higher average returnin each sample year and the range between theextreme outcomes in each subsample varies consid-erably on a year-to-year basis. On the other hand,there does not seem to be a discernable difference inthe spending policies of the two institutional types.

Finally, we also collected annual return data forbenchmark indices representing the investable asset



Tab

le1

Ove

rvie

wof

the

educ

atio

nale

ndow

men

tfun

dun

iver

se.

AU

M($

mill

.)R

etur

nPa

yout

No.

Year

obs

Tota

lM

ean

Med

ian

Std

Min

Max

Mea

nM

edia

nSt

dM

inM

axM

ean

Med

ian

Std

Min

Max

Pane

lA.F

ulls

ampl

e20

0570

924

9297

.29

352.

6177

.43

1294

.33

1.26

2547

3.72

9.16

9.00

3.29

−11.

4022

.20

4.78

4.85

1.40

0.24

17.1

020

0470

522

8466

.66

324.

5371

.88

1138

.76

1.92

2214

3.65

15.0

215

.80

4.42

−0.6

032

.10

4.88

5.00

1.62

0.00

18.4

020

0366

519

6111

.70

294.

9072

.42

1003

.83

0.32

1884

9.49

2.69

2.60

3.67

−14.

7028

.10

5.21

5.00

1.51

0.10

22.0

020

0253

517

6545

.39

336.

9288

.89

1053

.23

0.16

1716

9.76

−5.9

5−6

.20

4.23

−19.

8010

.10

5.22

5.00

1.54

0.30

22.0

020

0156

821

8185

.98

393.

1310

6.24

1172

.01

1.15

1795

0.84

−3.2

4−3

.55

6.31

−26.

9024

.80

5.17

5.00

1.43

0.30

14.2

020

0050

722

2928

.50

462.

5111

9.99

1373

.86

6.19

1884

4.34

12.6

710

.70

10.1

2−1

2.20

58.8

04.

975.

001.

400.

1013

.30

1999

467

1817

81.6

141

0.34

132.

4710

84.9

97.

1914

256.

0010

.79

10.5

04.

71−1

5.80

29.3

04.

825.

001.

240.

1013

.00

1998

445

1596

20.9

337

1.21

114.

1599

1.14

5.17

1301

9.74

18.0

618

.10

3.97

3.70

31.8

04.

694.

701.

210.

3014

.70

1997

456

1451

22.9

732

8.33

99.3

085

8.51

5.78

1091

9.67

20.1

019

.90

4.44

6.80

46.9

04.

804.

601.

770.

3024

.00

1996

405

1126

03.9

428

2.22

81.2

373

8.88

3.65

8811

.78

17.2

517

.20

4.02

0.80

39.0

04.

844.

801.

300.

1012

.80

1995

422

9941

6.70

236.

1472

.08

600.

900.

0770

45.8

615

.06

14.8

04.

150.

5037

.90

4.95

5.00

1.48

0.10

15.0

019

9437

570

354.

3118

9.12

61.2

340

6.42

0.47

3529

.00

3.25

3.20

4.35

−13.

0043

.00

5.24

5.00

1.75

1.80

18.2

019

9339

476

838.

2119

6.02

60.3

646

8.95

1.46

5778

.26

13.1

513

.30

4.28

−2.8

036

.30

n.a.

n.a.

n.a.

n.a.

n.a.

1992

318

6223

6.12

196.

3353

.66

533.

941.

7353

15.4

713

.00

13.0

04.

941.

1065

.00

n.a.

n.a.

n.a.

n.a.

n.a.

1991

328

5602

3.55

171.

8551

.67

437.

943.

3951

18.1

27.

257.

104.

89−1

4.40

53.0

0n.

a.n.

a.n.

a.n.

a.n.

a.19

9029

851

420.

1617

4.31

58.8

941

0.97

2.01

4653

.23

10.0

29.

556.

01−1

0.30

75.0

0n.

a.n.

a.n.

a.n.

a.n.

a.19

8928

145

818.

3916

5.41

56.8

739

0.98

1.87

4478

.98

13.5

813

.40

4.65

−6.2

037

.40

n.a.

n.a.

n.a.

n.a.

n.a.

1988

262

4946

7.06

157.

0447

.64

381.

401.

6541

55.7

81.

260.

704.

78−1

4.70

17.4

0n.

a.n.

a.n.

a.n.

a.n.

a.19

8727

645

222.

9815

2.78

46.3

836

0.68

1.54

4018

.27

13.6

313

.82

5.69

−6.8

831

.89

n.a.

n.a.

n.a.

n.a.

n.a.

1986

258

3931

7.86

150.

6446

.29

354.

881.

1034

35.0

127

.17

27.1

66.

725.

2452

.46

n.a.

n.a.

n.a.

n.a.

n.a.

1985

280

3151

1.82

112.

5431

.19

295.

301.

0229

27.2

025

.61

25.9

56.

961.

2552

.12

n.a.

n.a.

n.a.

n.a.

n.a.

1984

200

2539

9.06

127.

0040

.12

289.

120.

9724

86.3

0−1

.40

−2.2

415

.12

−17.

0020

9.00

n.a.

n.a.

n.a.

n.a.

n.a.



Tab

le1

(Con

tinue

d)

AU

M($

mill

.)R

etur

nPa

yout

No.

Year

obs

Tota

lM

ean

Med

ian

Std

Min

Max

Mea

nM

edia

nSt

dM

inM

axM

ean

Med

ian

Std

Min

Max

Pane

lB.P

riva

teen

dow

men

ts20

0553

519

6009

.71

367.

0677

.96

1430

.50

1.26

2547

3.72

9.25

9.00

3.19

−1.0

022

.20

4.79

4.81

1.43

0.24

17.1

020

0453

117

9359

.61

338.

4172

.79

1258

.52

1.92

2214

3.65

15.0

715

.80

4.30

−0.6

025

.30

4.86

5.00

1.56

0.00

18.4

020

0350

615

4276

.44

304.

8972

.93

1097

.94

1.58

1884

9.49

2.78

2.80

3.65

−14.

7028

.10

5.18

5.00

1.53

0.10

22.0

020

0240

613

8023

.88

345.

9287

.95

1147

.49

1.17

1716

9.76

−5.9

8−6

.30

4.15

−19.

809.

705.

165.

001.

341.

5013

.00

2001

438

1749

38.8

340

8.74

104.

5412

76.7

81.

1517

950.

84−3

.23

−3.5

06.

17−2

6.90

14.3

05.

105.

001.

301.

5014

.20

2000

392

1851

27.6

849

4.99

121.

3815

06.8

66.

1918

844.

3412

.60

10.5

010

.38

−9.0

058

.80

4.91

5.00

1.35

0.10

13.3

019

9936

315

1356

.26

436.

1913

0.10

1191

.00

7.19

1425

6.00

10.7

810

.50

4.81

−15.

8029

.30

4.77

4.80

1.20

0.10

13.0

019

9834

912

8724

.41

379.

7211

0.85

1069

.44

5.17

1301

9.74

18.0

118

.10

3.98

3.70

31.8

04.

664.

601.

212.

0014

.70

1997

354

1177

36.2

834

2.26

97.6

493

2.95

7.50

1091

9.67

20.1

420

.05

4.48

6.80

46.9

04.

814.

601.

882.

1024

.00

1996

320

9273

2.84

293.

4683

.20

796.

173.

6588

11.7

817

.25

17.1

54.

030.

8039

.00

4.87

4.80

1.30

2.30

12.8

019

9533

080

851.

0224

5.75

72.5

665

1.06

0.07

7045

.86

15.1

114

.85

4.17

0.50

37.9

04.

975.

001.

561.

0015

.00

1994

295

5744

4.37

196.

7360

.96

429.

880.

4735

29.0

03.

363.

104.

61−1

3.00

43.0

05.

165.

001.

781.

8018

.20

1993

310

6340

7.14

205.

8760

.18

509.

772.

2057

78.2

613

.10

13.2

54.

43−2

.80

36.3

0n.

a.n.

a.n.

a.n.

a.n.

a.19

9224

951

462.

6020

6.68

53.6

658

4.29

1.73

5315

.47

13.0

913

.00

5.27

1.50

65.0

0n.

a.n.

a.n.

a.n.

a.n.

a.19

9126

347

809.

0018

2.48

51.6

547

9.09

3.39

5118

.12

7.36

7.20

5.08

−14.

4053

.00

n.a.

n.a.

n.a.

n.a.

n.a.

1990

228

4127

6.38

183.

4560

.72

450.

132.

0146

53.2

310

.25

9.80

6.40

−10.

3075

.00

n.a.

n.a.

n.a.

n.a.

n.a.

1989

221

3913

5.75

178.

7056

.87

430.

851.

8744

78.9

813

.79

13.6

04.

78−6

.20

37.4

0n.

a.n.

a.n.

a.n.

a.n.

a.



Tab

le1

(Con

tinue

d)

AU

M($

mill

.)R

etur

nPa

yout

No.

Year

obs

Tota

lM

ean

Med

ian

Std

Min

Max

Mea

nM

edia

nSt

dM

inM

axM

ean

Med

ian

Std

Min

Max

Pane

lC.P

ublic

endo

wm

ents

2005

174

5328

7.58

308.

0272

.40

729.

821.

7754

33.0

98.

898.

803.

60−1

1.40

18.1

04.

754.

921.

290.

4211

.19

2004

174

4910

7.04

282.

2266

.70

651.

152.

9049

51.6

514

.88

15.6

54.

770.

2032

.10

4.93

5.00

1.78

0.00

12.0

020

0315

941

835.

2526

3.11

68.1

361

6.83

0.32

4527

.90

2.42

2.40

3.75

−12.

1014

.10

5.34

5.00

1.43

1.00

11.2

020

0212

938

521.

5130

8.17

95.8

367

1.85

0.16

4343

.72

−5.8

7−5

.90

4.49

−17.

8010

.10

5.40

5.00

2.03

0.30

22.0

020

0113

043

247.

1534

0.53

113.

8771

5.19

7.59

4857

.27

−3.2

8−3

.85

6.82

−20.

5024

.80

5.40

5.00

1.76

0.30

12.8

020

0011

537

800.

8235

0.01

117.

6174

3.66

7.05

5798

.53

12.9

111

.40

9.18

−12.

2047

.30

5.17

5.00

1.55

0.30

11.9

019

9910

430

425.

3431

6.93

140.

4754

7.26

8.22

4445

.60

10.8

310

.55

4.37

−4.3

022

.00

5.01

5.00

1.37

0.30

11.2

019

9896

3089

6.53

339.

5213

2.05

621.

337.

9337

87.8

818

.24

17.9

53.

979.

0031

.60

4.82

5.00

1.25

0.30

10.0

019

9710

227

386.

6927

9.46

104.

5251

9.56

5.78

3216

.65

19.9

319

.65

4.29

9.20

36.1

04.

754.

701.

340.

309.

1019

9685

1987

1.10

239.

4180

.39

460.

914.

1226

35.5

517

.23

17.2

03.

982.

4028

.90

4.73

4.85

1.29

0.10

8.70

1995

9218

565.

6820

1.80

67.1

237

0.35

3.31

2190

.59

14.8

814

.75

4.07

1.30

30.2

04.

895.

001.

150.

107.

3019

9480

1290

9.94

161.

3762

.24

306.

642.

7219

18.1

52.

863.

303.

20−5

.80

13.0

05.

575.

301.

603.

8013

.00

1993

8413

431.

0715

9.89

63.5

427

0.39

1.46

1846

.60

13.3

013

.60

3.72

4.00

22.4

0n.

a.n.

a.n.

a.n.

a.n.

a.19

9269

1077

3.52

158.

4353

.86

281.

154.

2416

83.0

112

.66

13.0

03.

501.

1019

.80

n.a.

n.a.

n.a.

n.a.

n.a.

1991

6582

14.5

512

8.35

52.2

019

0.10

4.29

1151

.32

6.81

6.90

4.02

−10.

9015

.80

n.a.

n.a.

n.a.

n.a.

n.a.

1990

7010

143.

7814

4.91

52.5

924

6.37

3.07

1494

.93

9.29

9.15

4.47

1.20

26.2

0n.

a.n.

a.n.

a.n.

a.n.

a.19

8960

6682

.64

115.

2255

.20

164.

232.

6997

3.70

12.8

312

.65

4.09

1.30

22.1

0n.

a.n.

a.n.

a.n.

a.n.

a.

Thi

sta

ble

repo

rts

sum

mar

yst

atis

tics

for

the

sam

ple

ofen

dow

men

tsin

the

NA

CU

BO

endo

wm

ent

stud

yda

taba

se.I

nclu

ded

are

annu

alda

taon

the

num

ber

ofpa

rtic

ipat

ing

fund

s,as

sets

unde

rman

agem

ent(

AU

M),

net-

of-f

eein

vest

men

tper

form

ance

,and

endo

wm

entp

ayou

t.Pa

nelA

lists

sum

mar

yst

atis

tics

fort

heen

tire

univ

erse

,whi

lePa

nels

Ban

dC

pres

ent

data

for

the

fund

sof

priv

ate

and

publ

icsc

hool

s,re

spec

tive

ly.



Table 2 Benchmark indices.

Summary statistics

Asset class Benchmark index Mean Std Median SR

1. U.S. Equity CRSP Value Weighted 11.47 13.09 13.13 0.562. Non-U.S. Equity MSCI World (Excl. US) 4.46 13.31 5.82 0.023. U.S. Fixed-Income Lehman Bond Aggregate 8.04 4.41 8.64 0.884. Non-U.S. Fixed-Income Salomon Brothers Non-U.S. Bond Index 7.89 9.08 7.60 0.415. Public Real Estate NAREIT 13.27 12.79 9.06 0.716. Private Real Estate NCREIF 7.81 6.55 8.07 0.567. Hedge Funds HFRI-all fund Composite 14.31 7.61 13.09 1.348. Venture Capital Cambridge Associate VC index 26.59 56.20 17.42 0.409. Private Equity Cambridge Associate PE index 14.76 13.74 15.38 0.77

10. Natural Resources AMEX Oil (before 1992), GSCI (after 1992) 7.02 16.87 1.33 0.1711. Other Investments — — — — —12. Cash 30-day U.S. T-Bill 4.14 1.92 4.73 0.00

This table lists the indices used as benchmarks for the asset classes represented in the NACUBO endowment study sample. Summary statisticsare listed in percent per annum. Data for all of the indices are obtained from Bloomberg, Cambridge Associates and the University of TexasInvestment Management Company.

classes covered by the NACUBO endowment sur-veys. We define 12 distinct asset classes: U.S. Equity,Non-U.S. Equity, U.S. Fixed-Income, Non-U.S.Fixed-Income, Public Real Estate, Private RealEstate, Hedge Funds, Venture Capital, PrivateEquity, Natural Resources, Other Investments, andCash.9 The benchmark definitions for these cat-egories are shown in Table 2, which also listsseveral summary statistics for the annual invest-ment performance they produced during the sampleperiod.10

4 Are endowments exhausting their activemanagement skills?: Empirical analysis

4.1 Decomposing endowment fund performance

Before proceeding with our assessment of whetherendowment managers adopt an optimal mixture ofactive and risk exposures in their portfolios, it isfirst useful to examine the investment performancein the fund sample over time. That is, are thesefunds capable of producing the consistent levels of

alpha that would justify any active risk allocation?We address this question by analyzing the realizedvalues of the annual return components producedby our endowment universe. Table 3 summarizestwo dimensions of the annual peer group compar-isons for the total return (R) as well as the passive(RB) and active elements (RT and RS). Specifi-cally, the display reports (i) the mean return foreach component and (ii) the median return andinter-quartile range (i.e., the difference in returnsproduced by the endowments falling at the 75thand 25th percentiles of a particular distribution)for the overall return and the sum of the two activecomponents.11

The reported findings support three primary con-clusions about endowment fund performance overthis period. First, as shown in the second and fifthcolumns of the display, both the overall returnsand passive return components that the endow-ment universe generated are considerable. The totalreturn averaged just under 10% per annum overthe entire time frame, with positive mean returns



Table 3 Decomposition of historical endowment fund returns.

Total return ActivePolicy Tactical Selection

Year Mean Median p75–p25 mean mean mean Mean Median p75–p25

2005 9.67 9.50 3.40 9.93 −0.52 0.26 −0.26 −0.13 2.782004 15.63 16.30 4.15 15.70 −0.44 0.37 −0.06 0.25 3.232003 2.37 2.40 3.20 3.08 0.35 −1.06 −0.71 −0.55 3.272002 −6.13 −6.30 4.45 −8.26 0.70 1.43 2.13 1.99 4.432001 −3.42 −3.70 6.90 −8.63 −0.38 5.59 5.21 5.22 7.062000 13.02 11.00 10.15 11.70 0.90 0.42 1.32 0.50 8.151999 10.85 10.60 5.38 11.77 0.28 −1.19 −0.92 −0.98 5.701998 18.09 18.10 4.30 17.91 0.12 0.06 0.18 0.21 4.291997 20.14 20.10 5.00 17.46 1.10 1.58 2.68 2.35 5.331996 17.35 17.20 3.80 14.99 0.28 2.08 2.36 2.13 3.741995 15.26 15.00 4.10 16.51 −0.68 −0.57 −1.26 −1.48 4.551994 3.39 3.20 3.15 1.14 0.16 2.09 2.25 2.01 2.981993 13.46 13.60 4.80 13.13 0.29 0.04 0.33 0.59 4.921992 13.00 13.00 3.08 12.07 0.12 0.81 0.93 0.75 3.861991 6.77 6.80 5.00 7.05 0.29 −0.57 −0.28 −0.07 4.15

Grand mean 9.96 — — 9.04 0.17 0.76 0.93 — —

This table presents mean annual values for the various components of endowment fund returns from the NACUBO database as wellas their grand time series means over the entire sample period. “Total Return” represents the overall returns reported to NACUBO.“Policy” are the average returns of the policy portfolios. “Active” represents the returns from active investment strategies, whichinclude the “Tactical” and “Selection” categories. The exhibit also reports the annual inter-quartile (i.e., p75–p25) return spreadsfor the Total and Active return categories. Rounding accounts for any apparent discrepancies across categories.

reported in 13 of the 15 years represented. Further,the average passive return component was 9.04%for the whole period, which amounts to more than90% (i.e., 9.04/9.96) of the mean total return. Ofcourse, this finding is consistent with the conclu-sion of both Brinson et al (1986) and Ibbotson andKaplan (2000) who showed that, over time, thestrategic asset allocation decision is the most impor-tant determinant of the investment performance forany given fund.

The second thing to note fromTable 3 is that there isa strong indication that endowment fund managersdo indeed possess tangible alpha-generating skills.This is the more relevant finding for the presentinvestigation because in the absence of active man-agement skill the question of the appropriate way

to combine passive and active risk exposures inthe portfolio becomes moot. From the third-to-lastcolumn of the exhibit, the overall mean alpha gen-erated by the endowment fund sample is almost100 basis points per year, with a range across timeof −1.26% to 5.21%. (Comparable values arereported in the penultimate column for the medianfund.) Additionally, notice from the Tactical andSelection return columns that the latter componentis the larger and more reliable determinant of thetotal active returns produced by the fund universe.In fact, when viewed over the entire sample period,security selection skills accounted for roughly 82%(i.e., 0.76/0.93) of the observed alpha.

Finally, while the mean alpha component dataindicate that the average endowment manager is a



genuinely skillful active investor, it is not necessarilythe case that all of the endowments in the sam-ple are equally capable in this regard. In fact, fromthe size of the reported inter-quartile ranges for thetotal and active return components, we can inferthat there is a large degree of cross-sectional vari-ation in the management abilities represented inthe endowment universe. In particular, the last col-umn in the display shows that, in a given year, thespread in active management performance betweenthe upper and lower quartiles of the fund distribu-tion can exceed 800 basis points, an amount that isabout nine times the size of overall mean alpha itself!Further, this inter-quartile active return spread isnever less than 278 basis points (i.e., in 2005).Finally, juxtaposing the values of the Total Returnand Active inter-quartile ranges, it is clear that theactive return component is the primary determinantof deviations in the overall investment performanceacross the sample. Thus, it is likely that there areimportant fund-level characteristics that distinguishthe investment decision-making abilities of thesemanagers.

4.2 Determining the optimal mix of active andpassive risks: Full sample results

From the discussion in Section 2.3, the problemof finding the optimal λ can be thought of asa standard maximization problem with only twocomposite assets: R and RB. Suppose, for instance,that an endowment fund’s objective is to maximizeits Sharpe ratio. If such an endowment is alreadychoosing λ optimally, then, by construction, thesolution in Eq. (8) will deliver an optimal weightλ = 1. The same holds true if the endowment max-imizes its Sortino ratio instead and so the solutionof the optimization problem for λ in either Eq. (8)or Eq. (9) provides us with a natural metric withwhich to analyze whether endowments are alreadyoptimally exploiting their active investment abilities(i.e., adopting an appropriate active risk alloca-tion). An optimal value of λ > 1 implies that the

fund is not fully exploiting its active managementcapabilities, and may have benefited from tiltingits portfolio away from the benchmark indices andmore toward its specific tactical asset class weightsand individual security holdings.

To establish empirically the optimal exposure λ toactive risk for each fund in our whole sample, weestimated the moments of the two composite assetsR and RB by using the entire observed NACUBOendowment fund return time series from 1989 to2005. We then solved separately for the λ valuesin Eqs. (8) and (9) that maximized the Sharpe ratioand the Sortino ratio, respectively.12

Table 4 summarizes our findings. When the fund’sobjective is to maximize the Sharpe ratio, Panel Ashows that the weights λ resulting from the max-imization in Eq. (8) have a cross-sectional meanof 1.29 (t -statistic of 4.40) and a median of 1.17.As per our previous discussion, this finding can beinterpreted as follows: It would have been optimalover this investment horizon for the typical endow-ment to sell 29% of its passive benchmark portfolioand hold 29% more of its actual managed portfolio.Maximizing Eq. (9) for the Sortino ratio produces asimilar conclusion but with even stronger findings;the resulting cross-sectional average of the distribu-tion of estimated λ coefficients is 1.52 (t -statistic of6.20) and the median is 1.44.13

Because these initial findings are based on an in-sample analysis, it is reasonable to ask whether anendowment could still exploit such a strategy afterlearning about the extent of its active managementabilities. Thus, a more interesting exercise is to ana-lyze the out-of-sample consequences for a fund thatdecides to alter the active–passive risk profile in itsportfolio based on the information revealed by thesein-sample findings. To examine this possibility, weseparated the sample into two distinct sub-periodsand used the earlier of the two (i.e., from 1989until 1997) to generate estimates of the moments ofthe current (R) and the benchmark (RB) portfolios



Table 4 Optimal active–passive allocation.

Optimization criterion Mean λ Median λ

Panel A: In-sample values of λ

Sharpe ratio 1.29 1.17(t -stat.) (4.40) (—)

Sortino ratio 1.52 1.44(t -stat.) (6.20) (—)

Average AveragePortfolio Sharpe ratio Sortino ratio

Panel B: Out-of-sample performanceWithout swap 0.3012 0.5228With swap 0.3533 0.6307

Difference 0.0521 0.1079(t -stat.) (2.65) (4.05)

Average AveragePortfolio standard deviation downside deviation

Panel C: Out-of-sample risk changeWithout swap 10.21% 6.68%With swap 11.23% 6.92%

Difference 1.01% 0.24%(t -stat.) (3.15) (3.71)

Panel A reports the scale parameter λ that maximizes the in-sample Sharpe or reward-to-downside-risk (Sortino) ratio, computed as described in Section 4. Panel B reports theout-of-sample Sharpe and Sortino ratios obtained by implementing during the period 1998–2005 the swap between the active and passive parts of the portfolio according to the value of thescale λ determined in the period 1989–1997. Panel C reports the out-of-sample relevant riskstatistics (standard deviation and downside deviation) calculated over the period 1998–2005when the swap determined during the period 1989–1997 is implemented.

in the same manner as before. With these samplestatistics, we calculated the optimal rescaling factorλ as of the end of 1997, for both the Sharpe andSortino decision criteria. We then assumed that theendowments for which the optimizer recommendsa weight of λ > 1 in 1997 (210 out of 339 endow-ments in 1997 for the Sharpe ratio case and 247 forthe Sortino ratio case) actually did rescale their port-folio accordingly for the second half of the sample(i.e., from 1998 to 2005). Endowments for whichλ < 1 were assumed to not implement the swap.

The results of this out-of-sample analysis arereported in Panel B of Table 4. When the set ofλ values is obtained from maximizing Eq. (8) dur-ing the period 1989–1997, the implied strategy ofenhancing active risk in the fund by holding λ ofthe actual portfolio results in an average increase inthe endowment’s Sharpe ratio of 0.0521 (t -statisticof 2.65) over the period 1998–2005 (i.e., froma ratio of 0.3012 without the swap to 0.3533with the swap). Further, the median Sharpe ratioincrease is 0.0271 (z-statistic of 2.65).14 Similarly,



when the λ estimates are obtained from Eq. (9),the resulting increase in the average Sortino ratiofor the endowments that follow the strategy ofenhancing the active part of their portfolio is equalto 0.1079 (t -statistic of 4.05), from 0.5228 to0.6307; the median Sortino ratio increase is 0.0672(z-statistic of 2.81). These findings suggest quitestrongly that: (i) the typical endowment fund didnot take enough active risk over the sample period,and (ii) increasing the scale of the same alpha-generating skills it was already deploying wouldhave significantly increased the level of the portfo-lio’s risk-adjusted performance. With respect to thelatter point, it is important to mention that scalingup alpha-generating skills does not necessarily implyan increased level of trading; the endowment couldsimply hold more of its overall portfolio in positionsthat are less correlated with the benchmark.

Another potentially important concern related tothis conclusion is the effect that the swap has on theoverall portfolio risk. In Panel C ofTable 4 we reportthe changes in overall portfolio risk following swapimplementation, using both the standard deviationor downside deviation of returns as volatility mea-sures. The main observation from these findings isthat the level of overall portfolio risk does indeedincrease somewhat if the swap is implemented andthat the increase is statistically—if not necessarilyeconomically—significant. When the swap result-ing from Eq. (8) is implemented, the standarddeviation of the portfolio increases from 10.21% to11.23%. (We get a similar result when the swap ofimplied by Eq. (9) is implemented.) Thus, despitethe fact that this increase in portfolio volatility ismore than offset by the simultaneous increase inreturn, there may well be some endowments whichwould be allowed to implement the swap only ifsuch a risk increase is acceptable to their investmentcommittees.

Finally, it is also instructive to address the fol-lowing question: Which endowments are likely to

see an increase in overall fund volatility followingthe implementation of the recommended active–passive risk swap? Table 5 attempts to answer thisissue. Specifically, the display reports the estimatedparameters of a cross-sectional regression of thevolatility increases associated with the swap on sev-eral endowment-specific characteristics, includingthe correlation between the actual and benchmarkportfolio returns, payout level, AUM, allocation tovarious broad asset classes, and the level of totalfund risk prior to swap implementation. The onlycharacteristic related to the post-swap risk increasethat proves to be statistically significant for both riskmeasures is the level of fund risk prior to the swapimplementation. This is intuitive: If an endow-ment held an overall riskier portfolio before theswap, shorting the benchmark to buy even moreof it should clearly translate into a post-swap port-folio with a larger risk increase relative to the restof funds. Finally, it is interesting to note that, con-trary to conventional wisdom, scaling up the activerisk exposure in a fund with a larger allocation tothe alternative asset classes does not result in anincreased overall level of volatility relative to endow-ments with a smaller allocation to hedges funds andprivate equity.

4.3 Optimal active–passive risk combinations:Cross-sectional results

The preceding section provides strong evidence sup-porting the notion that the typical endowment fundunderutilizes its alpha-generating ability and couldsignificantly improve its performance by scaling upthe active risk exposure in its portfolio. What istrue on average, however, is not necessarily trueacross the entire sample. In fact, the descriptivestatistics on the fund universe presented in Table 1suggest that there are at least three characteristics—the AUM of the fund, the spending rate of theendowment, and whether sponsoring institutionis a public or a private school—that could have amaterial impact on the investment decisions made



Tab

le5

Cro

ss-s

ecti

onal

dete

rmin

ants

ofri

skch

ange

follo

win

gsw

apim

plem

enta

tion

.

Dep

ende

ntC

orre

lati

onlo

gA

lloca

tion

toA

lloca

tion

toA

lloca

tion

toPa

stA

djus

ted

vari

able

Inte

rcep

t(a

ctua

l,po

licy)

Payo

utA

UM

alte

rnat

ives

fixed

inco

me

equi

tyri

skR

-squ

are

�(s

tand

ard

devi

atio

n)−0

.109

80.

0798

0.00

000.

0005

0.01

260.

0392

0.01

550.

5274

t-st

at.

−1.7

477

2.83

320.

3558

0.24

550.

1445

0.64

890.

2553

3.00

306.

93%

�(d

owns

ide

devi

atio

n)−0

.040

80.

0053

−0.0

000

−0.0

013

0.09

140.

0667

0.03

190.

3576

t-st

at.

−0.8

841

0.31

62−0

.853

4−0

.906

51.

5174

1.51

970.

7107

1.84

652.

65%

Thi

stab

leco

nsid

erst

heef

fect

sofs

wap

impl

emen

tati

onon

fund

risk

byre

port

ing

the

esti

mat

edpa

ram

eter

sofa

cros

s-se

ctio

nalr

egre

ssio

nof

the

chan

gein

post

-sw

apen

dow

men

tri

skon

the

follo

win

gva

riab

les:

the

corr

elat

ion

betw

een

actu

alan

dpo

licy

retu

rns,

payo

utle

vel,

endo

wm

entA

UM

,allo

cati

onto

alte

rnat

ive

asse

tcla

sses

,allo

cati

onto

fixed

-inc

ome,

allo

cati

onto

publ

iceq

uity

,and

pre-

swap

risk

leve

l.t-

Stat

isti

csar

ere

port

edbe

neat

hea

chre

spec

tive

coef

ficie

ntes

tim

ate.



by the fund manager. In this section, we extendour previous analysis on the optimality of endow-ment active–passive risk exposures by looking at thefindings on a cross-sectional basis.

Table 6 repeats our in-sample and out-of-sample cal-culations for the rescaling parameter λ based on thefollowing ways of dividing the endowment universeinto tractable subsamples: The largest and small-est quartiles based on fund AUM (Panel A), thelargest and smallest quartiles based on endowmentspending rate (Panel B), and public versus privateendowments (Panel C). In particular, for each ofthe respective divisions of the endowment universe,we calculate the average λ value over the entire sam-ple period, as well as the change in the Sharpe andSortino ratios during the second half of the periodthat would accompany the implementation of anactive–passive swap based on information from thefirst half. To get a better sense of the pervasiveness oforiginal results, we then compute the cross-sectionalvariation in those λ and objective function statis-tics, along with the t -statistics affiliated with thosedifferences.

The results tell a mixed story about the importanceof characteristic variability across the endowmentsample with respect to the active–passive risk deci-sion. On one hand, Panel A1 of Table 6 indicatesthat there are significant differences in the rescalingparameters estimated for large and small endow-ments. For instance, using the Sharpe ratio torepresent the risk-adjusted objective, the averageλ coefficient for the quartile of the fund samplewith the largest AUM levels is almost twice thesize of that for the lowest-AUM funds (i.e., 1.93versus 1.07). There are two interesting aspects to thisfinding. First, it suggests that to achieve its optimalcombination of exposures, the average large endow-ment should virtually double its current portfolio byshorting almost all of its passive position. The out-of-sample performance metrics in Panel A2 showthat affecting this adjustment would dramatically

increase both the Sharpe and Sortino ratios for thelarge endowment, albeit at marginal levels of statis-tical reliability. Of course, it is highly unlikely thatany fund would be allowed the freedom to actuallyimplement a swap that requires such an extreme useof leverage; this is a topic that we consider in moredetail in the next section. Nevertheless, the find-ings underscore the extent to which the managersof large endowment funds are not deploying theirfull active management skills.15

The second interesting aspect of the results shownin Panel A is that the estimated λ levels for theaverage small endowment are not significantly dif-ferent from 1.00. There is no appreciable increasein the performance statistics when the implied swaptransaction is implemented by these small-fundmanagers (e.g., the pre-swap and post-swap Sharperatios are 0.2661 and 0.2914, respectively, yield-ing an insignificant increase of 0.0253). Taken atface value, this indicates that small endowmentsare fully utilizing their alpha-generating talents andthese funds would gain very little by scaling up theircurrent positions in exchange for the passive port-folio. What is also likely to be true, however, isthat small funds have less access to active invest-ment strategies in the first place, meaning that theyhave less to “leave on the table” when making theiractive–passive allocation decisions. This interpreta-tion is consistent with the result reported by Brownet al. (2009) that large endowments invest aboutfive-to-ten times more of their proportional assetsin hedge funds and private equity than do smallendowments.

In contrast with these cross-sectional AUM find-ings, there is no apparent difference in the λ

and objective function estimates for the spendingrate or public–private subsamples. The coefficientsreported in both Panels B and C of the display showthat there are no statistically meaningful differencesin these values for either characteristic. That is,despite the heterogeneity documented across the



Table 6 Optimal active–passive allocation for different endowment types.

Panel A: Large versus small AUM endowments

Panel A1: In-sample values of λ

Average λOptimizationcriterion q4-Large q1-Small q4–q1

Sharpe ratio 1.93 1.07 0.86(t -stat.) (7.41) (0.48) (4.53)

Sortino ratio 2.29 1.26 1.03(t -stat.) (8.84) (1.36) (4.30)

Panel A2: Out-of-sample performance

Average Sharpe ratio Average Sortino ratio

Portfolio q4-Large q1-Small q4–q1 q4-Large q1-Small q4–q1

Without swap 0.3979 0.2661 0.1318 0.7672 0.3960 0.3712With swap 0.4763 0.2914 0.1848 0.9780 0.4597 0.5184

Difference 0.0783 0.0253 0.0531 0.2108 0.0636 0.1472(t -stat.) (1.06) (0.25) (0.90) (1.63) (0.46) (1.87)

Panel B: Large payout versus small payout endowments

Panel B1: In-sample values of λ

Average λOptimizationcriterion q4-Large q1-Small q4–q1

Sharpe ratio 1.25 1.22 0.03(t -stat.) (3.50) (1.97) (0.23)

Sortino ratio 1.56 1.66 −0.10(t -stat.) (6.39) (4.81) (−0.60)

Panel B2: Out-of-sample performance


Portfolio q4-Large q1-Small q4–q1 q4-Large q1-Small q4–q1

Without swap 0.3076 0.3181 −0.0105 0.5417 0.6126 −0.0708With swap 0.3657 0.4274 −0.0617 0.6615 0.8048 −0.1433

Difference 0.0581 0.1093 −0.0512 0.1197 0.1922 −0.0725(t -stat.) (1.08) (0.91) (−1.02) (1.54) (1.04) (−1.11)



Table 6 (Continued)

Panel C: Public versus private endowments

Panel C1: In-sample values of λ

Average λOptimizationcriterion Private Public Private–Public

Sharpe ratio 1.27 1.20 0.07(t-stat.) (3.73) (1.38) (0.45)

Sortino ratio 1.54 1.51 0.03(t -stat.) (6.00) (2.59) (0.13)

Panel C2: Out-of-sample performance


Portfolio Private Public Private–Public Private Public Private–Public

Without swap 0.3044 0.2890 0.0154 0.5526 0.4310 0.1216With swap 0.3676 0.3084 0.0592 0.6717 0.5015 0.1702

Difference 0.0632 0.0194 0.0438 0.1192 0.0705 0.0486(t -stat.) (1.13) (0.27) (0.94) (1.58) (0.51) (0.74)

Panel Al reports the scale parameters λ that maximize the in-sample Sharpe ratio and reward-to-downside-risk(Sortino) ratio, computed as described in Section 4, for the top quartile q4 and the bottom quartile q1 of assetsunder management. Panel A2 then reports the out-of-sample Sharpe and Sortino ratios obtained by implementingthe swap between the active and passive parts of the portfolio during the 1998–2005 period according to the valueof λ estimated over the period 1989–1997. Large and small endowments are compared based on their assets undermanagement at the time when λ is determined in 1997. Panel B1 reports the scale parameters λ that maximizesthe in-sample Sharpe and Sortino ratios for the top quartile q4 and the bottom quartile q1 of fund payout ratios.Panel B2 then reports the out-of-sample Sharpe and Sortino ratios obtained by implementing the swap betweenthe active and passive parts of the portfolio during the 1998–2005 period according to the value of λ estimatedover the period 1989–1997. Large and small payout endowments are compared based on their reported payoutratios at the time when λ is determined in 1997. Panel C1 reports the scale parameters λ that maximizes thein-sample Sharpe and Sortino ratios for private and public endowments. Panel C2 then reports the out-of-sampleSharpe and Sortino ratios obtained by implementing the swap between the active and passive parts of the portfolioduring the 1998–2005 period according to the value of λ estimated over the period 1989–1997.

endowment universe for these variables themselves,using them to sort the sample made no differ-ence in how the optimality of the active–passiverisk allocation findings should be interpreted. Forinstance, the average rescaling parameters for thehigh-spending (i.e., 1.25) and low-spending (i.e.,1.22) quartiles using the Sharpe ratio objective func-tion are not significantly different from one another.Further, both values are quite comparable to the

average λ level for the overall sample (i.e., 1.29).A strikingly similar pattern emerges when the uni-verse is divided into public and private institutionalsponsorship. Thus, while both the spending needsand the profile of the educational structure mightimpact other aspects of the endowment manage-ment process, they do not appear to influence ona cross-sectional basis the decision of how muchactive risk to take.



5 Can the active–passive swap strategy beimplemented?: Normative implications

Based on the analysis in the previous section, areasonable question to ask is: If endowment fundmanagers could improve their risk-adjusted perfor-mance by this active–passive swap, why are theynot already implementing the strategy? Broadlyspeaking, two answers are possible. First, it couldbe the case that the managers and sponsors ofthe various funds are unaware of—or otherwisefail to acknowledge—the active management skillsthat they possess within the organization. How-ever, given the historical track record of producingpositive alphas documented in Table 3, this seemsunlikely to be the case. Indeed, the incentive-basedperformance (i.e., bonus) portion of the contractsthat typify compensation arrangements in this partof the asset management industry is based on theexplicit premise that managers do have measurableactive management skills.

The second and more plausible reason that endow-ment fund managers appear to adopt a consistentlysuboptimal combination of active and passive riskexposures is that there are impediments preventingthem from implementing the superior alternative.Said differently, although fund managers recognizethat they are underutilizing their alpha-generatingtalents, they are prevented from altering the port-folio in a way that would correct the problem. Weconsider the normative implications of two possibleimpediments along these lines: Market scalabilitylimitations and agency problems that may existwithin the endowment fund complex.

The first potential problem that may be prevent-ing fund managers from utilizing their skills fullyis related to the fact that the strategies and port-folio positions responsible for the levels of alphawe observe in the sample are not easily scaledup in size. Such an obstacle would be consistentwith the evidence reported by Chen et al. (2004),who showed that active returns in the mutual

fund industry declined significantly as the sizeof the fund grew, particularly when the invest-ment involved transactions in small and illiquidsecurities.16 In fact, of the two sources of alpha—tactical strategies and security selection—the latterwould seem to be the most susceptible to thissort of “economy of scale” constraint; managershave a far deeper and more liquid array of choices(e.g., ETFs, index-linked derivatives) when imple-menting market timing trades. Thus, given ourearlier finding that the selection component is thedominant driver of total active endowment fundreturns, this limitation becomes a relevant factor.Clearly, endowments can only implement the rec-ommended active–passive swap to the extent thatreturns to the security selection component are notalready completely exhausted.

Assessing the direct impact that this scalabilityimpediment might have on the endowment uni-verse would require complete information aboutfund-level holdings, which is not available in ourdatabase. However, we presented findings in theprevious section that provide strong indirect evi-dence that this sort of market-driven liquidityconstraint does influence investment behavior inthe endowment industry. Recall from Panel A ofTable 6 that there was a tractable difference in howlarge and small funds managed their active–passiverisk exposures. In particular, it was the managers ofthe largest endowments who appeared to be deploy-ing only half of their active management skills,whereas small-fund managers were shown to beexploiting their full capabilities. This outcome isentirely consistent with the notion that large-fundmanagers are simply not able to scale up their alpha-generating investments to a greater extent (e.g., itmay not be possible for an endowment to increase a$25 million limited partnership position in a hedgefund vehicle to a $50 million position on a discre-tionary basis). Conversely, consistent with Brownet al. (2008), small-fund managers may be unableto access hedge fund partnerships in a profitable



manner because their due diligence and monitoringcosts are too high.

A second explanation for the apparent suboptimal-ity we observe in the way endowments manage theiractive–passive risk allocations may lie in the agencyproblems that exist in any principal-agent relation-ship. Besley and Prat (2003) as well as Walter(2004) have examined these conflicts in the spe-cific case of the asset management industry andhave identified three areas of friction that mightexist between boards and managers: The respon-sibility for monitoring the manager, risk exposuresand asset allocation decisions, and portfolio fundinglevels. They argued that if the decision-makers (i.e.,the portfolio managers) do not bear the full costof their actions, inefficient actions that adverselyimpact the position of the residual claimant (i.e.,the fund sponsors) can result. In fact, Brown et al.(1996) have shown the conditions in which fundmanagers alter the overall risk level of their port-folios in a manner consistent with an attempt tomaximize their personal compensation.

This potential for agency conflicts would cer-tainly seem to be present in the endowment fundindustry. One natural way for fund sponsors toaddress these problems is to design investment pol-icy statements that limit activities viewed as beingpotentially self-serving for managers (e.g., exces-sive risk-taking, the use of leverage) on an ex antebasis. For instance, in their examination of pol-icy constraints used by mutual funds, Almazanet al. (2004) demonstrated the relatively wide-spread use of these covenants, particularly withregard to leverage-related strategies. Further, theyalso showed that these restrictions were more likelyto be implemented in situations where directlymonitoring investment managers was more costlyto the investor. Thus, there may be cases in whichendowment portfolios are under-allocated to activerisk because the managers are simply not permittedby their boards to do otherwise.17

While the possibility that policy restrictions impedethe optimal use of active risk is certainly plausi-ble, there is nothing in the cross-sectional evidencein Panels B and C of Table 6 that substantiatesthis notion. Assuming that endowments which facedifferent investment problems (i.e., have differentspending rates) or are sponsored by different institu-tional types (i.e., private versus public universities)might also have different sets of agency conflicts,the estimated λ levels for the various subsamplesmight also be expected to differ. For example, boththe boards and managers of public school fundsare often under the direct scrutiny of a legislativebody, which might imply a tighter set of restrictionsthan those faced by an otherwise comparable pri-vate endowment. However, as discussed earlier, anysuch differences did not result in measurably differ-ent rescaling parameter estimates. Of course, ratherthan indicate that agency problems do not matter,these data might simply imply that the nature ofthese conflicts is effectively the same for all endow-ments, regardless of the distinct characteristics thatseem to separate the funds.

6 Concluding comments

Managers of multi-asset class portfolios—such aspension or endowment funds—attempt to enhancethe performance of their strategic benchmarkallocations by deploying their active investmentskills in two ways: Market timing and securityselection decisions. Thus, a fundamental choicethat every such manager must make involves theappropriate mixture of passive and active expo-sures to include in the portfolio, which is tanta-mount to deciding the appropriate combinationof active and passive risks to assume. Do man-agers make the right decision or do they tend toeither over- or underutilize their alpha-generatingtalents?

In this study, we have addressed this question intwo ways. First, building on the essential insight



of Treynor and Black (1973) that the active–passiverisk allocation decision can be viewed as a port-folio optimization problem, we present a simpleframework for determining what represents an opti-mal way of combining the two. Using an extensivedatabase consisting of the allocation decisions andinvestment performance of a large set of univer-sity endowment funds over the period from 1989to 2005, we then examine empirically the extentto which these managers exploit their value-addingcapabilities. We found that the average endowmentunderutilizes its active management skills by about30% and that it could have significantly increasedits risk-adjusted performance by shifting the exist-ing portfolio away from its passive benchmarks andtoward more of the active investment strategies andpositions it already holds. Further, we show thatthis tendency is not uniformly distributed across theentire endowment universe but most pronouncedfor funds with the largest amount of assets undermanagement. Thus, the answer to the central ques-tion that frames our investigation certainly appearsto be “no.”

The ancillary question that remains unanswered iswhy do endowment fund managers stop short ofusing their skills to the fullest potential? Ratherthan simply concluding that these funds are investedinefficiently, we considered two possible impedi-ments to implementing a superior combination ofactive and passive exposures. First, we argued thatit is quite likely that the alphas we documented forour sample are decreasing in scale, or are other-wise cost-ineffective to replicate on a larger basis.Second, we also suggested that rescaling the endow-ment portfolio in the manner recommended bythe implied swap strategy might require transac-tions and alter risk in a way that creates additionalagency problems in the fund organization. Whileour cross-sectional findings only support the for-mer explanation, both are plausible alternatives thatmerit further research.

Acknowledgements

The authors thank Jessica Shedd and the NationalAssociation of College and University BusinessOfficers for furnishing much of the data requiredby this study. We are also especially grateful for thecontributions of Lorenzo Garlappi and the comm-ents of several other colleagues, including SteveDimmock, Elroy Dimson, John Griffin, Bing Han,Jennifer Huang, Bruce Lehmann, Bing Liang, JeffKubik, Clemens Sialm, Laura Starks, Sheridan Tit-man, Raman Uppal, Bas Werker, and RobertoWessels. We have also benefitted from conversationswith a number of people involved in the endow-ment fund industry: Bob Boldt and Andrea Reed(Perella Weinberg), Gary Hill, Cathy Iberg, and UziYoeli (University of Texas Investment ManagementCompany), John Griswold (Commonfund), LarrySiegel (Ford Foundation), Jim Hille (TCU Endow-ment Fund), and Chance Burhance (InternationalFund Services). Finally, the comments of an anony-mous referee have improved the exposition in thestudy greatly. All errors in the paper are solely ourresponsibility.

Notes

1 The list of studies representing the risk budgeting litera-ture is lengthy and space limitations prohibit providing anexhaustive catalog here. However, Pearson (2002) offers anexcellent introduction and overview of the topic.

2 For instance, both the capital asset pricing model of Sharpe(1964) and the arbitrage pricing model of Ross (1976) arewell-known general models of risk and expected return inwhich systematic risk is rewarded but unsystematic risk isnot.

3 This relationship between risk-adjusted performance,manager skill, and the breadth of the investment oppor-tunities is called “the law of active management,” andit has been the basis for substantial additional analysis;see, for example, Qian and Hua (2004) and Clarke et al.(2006). Also, Alford et al. (2003) as well as Harlow andBrown (2006) have explored the empirical benefits andlimitations of attempting to identify superior active fundmanagers.



4 There is substantial survey evidence that decomposingreturns in this manner is, in fact, a reasonable way tocharacterize the endowment management process. Forinstance, Griswold (2008) shows that endowments investbetween 65 and 85 percent of the funds dedicated to aparticular asset class in actively managed portfolios, withthe remainder being passively indexed.

5 More recently, Lo (2008) has suggested an alternativemethod for decomposing active and passive portfolioreturns based on the manager’s ability to forecast futuremarket trends. As described below, however, this methodrequires data that were not available for this studyand hence that proposed method cannot be examinedherein.

6 Specifically, for a given λ we construct a time series of theswap portfolio returns Rt (λ) = λRt + (1−λ)RB

t . E [·] andstd[·] are the time series mean and standard deviation ofRt (λ) in excess of the risk-free rate Rf .

7 An excellent discussion of this topic can be found inIneichen (2007). Also, Harlow (1991) has demonstratedthe potential problems associated with the asset alloca-tion process when the potential for asymmetric returns isignored.

8 TIAA-CREF has administered the survey since 2000; from1988 to 1999, the survey was conducted in partner-ship with Cambridge Associates and before 1988 by theNACUBO Investment Committee.

9 Brown et al.’s (2009) study contains a more detaileddescription of these asset class designations as well as thor-ough examination of the specific allocation tendencies ofthe endowment fund universe in these classes.

10 Note that although the most popular U.S. Equity bench-mark used by endowments is the Russell 3000 index,we have used the Center for Research in Security Prices(CRSP) value-weighted market index portfolio instead.The reason is that many endowments also use the Wilshire5000 index as a benchmark for their U.S. Equity portfoliosand thus we chose to use a broad, representative index. Theresults we present in the following section are unchangedif the Russell 3000 index is used instead.

11 In our database, we do not observe the return ri,j,t foreach asset class that is necessary to calculate the value forRS directly from Eq. (4). However, this security selectionreturn component can be easily obtained by subtractingthe benchmark and market timing returns in Eqs. (2)and (3), respectively, from the total return in Eq. (1).Further, recall that for most years in the sample, the esti-mation of policy benchmark weights requires the use oflagged actual weights, which explains why 1991 is the first

year for which component returns are reported in Table 3despite the fact the sample period begins in 1989.

12 To minimize the effect of estimation error, we eliminatedfrom consideration both the outlier weights at the 5%levels in the cross-section before analyzing the propertiesof these distributions of λ values as well as all informationfor funds with fewer than five annual observations.

13 To assess the robustness of these findings with respectto estimation error in the means and covariance matrix,we have replicated this analysis using only subsamples ofendowments that have between six and fifteen years of dataavailable, as well as endowments for which quarterly dataare available. The results do not change and λ remainsstatistically significantly above 1.00 in all cases.

14 The distribution of the differences in the Sharpe ratiosbetween the rebalanced portfolio and the original oneranges from −0.50 to 0.69. The second highest Sharperatio differential is 0.68.

15 It is worth recalling from Section 2.1 that, in equilibrium,active management should be a zero-sum game. However,it need not be the case that active management withinthe endowment fund industry adds no net value. Indeed,it is unlikely that endowment funds trade only with oneanother; they may, for instance, have retail investors astheir most likely counterparties. Thus, there is no reasonto suspect that larger endowments are benefitting at theexpense of smaller funds.

16 Berk and Green (2004) have developed a formal theoreticalmodel for active portfolio management that makes a sim-ilar prediction when managerial talent is a scarce resourcethat is dissipated as the scale of fund’s operations increases.

17 At the time of his arrest in early 2009, Bernard Madoffwas the Chairman of the Board of Trustees for a privateuniversity with a $1.3 billion endowment fund. By the endof that year, the university reported that $110 million, orapproximately 8%, of their fund was lost due to fraudulentinvestment schemes. Thus, although the oversight andmonitoring provided by the Board via policy restrictionsand other measures are important safeguards in the overallmanagement process, they are hardly fool-proof tools thatcan be employed with perfect knowledge and foresight.

References

Alford, A., Jones, R., and Winkelmann, K. (2003). “Budget-ing Risk along the Active Risk Spectrum.” In Litterman,R. (ed.), Modern Investment Management: An EquilibriumApproach. Hoboken, NJ: John Wiley & Sons.



Almazan, A., Brown, K., Carlson, M., and Chapman, D.(2004). “Why Constrain Your Mutual Fund Manager?”Journal of Financial Economics 73, 289–321.

Berk, J. and Green, R. (2004). “Mutual Fund Flows and Per-formance in Rational Markets,” Journal of Political Economy112, 1269–1295.

Berkelaar, A., Kobor, A., and Tsumagari, M. (2006). “TheSense and Nonsense of Risk Budgeting,” Financial AnalystsJournal 62, 63–7.

Besley, T. and Prat, A. (2003). “Pension Fund Governanceand the Choice Between Defined Benefit and DefinedContribution Plans,” Working Paper, Institute for FiscalStudies.

Brinson, G., Hood, L., and Beebower, G. (1986). “Determi-nants of Portfolio Performance,” Financial Analysts Journal42, 39–48.

Brinson, G., Singer, B., and Beebower, G. (1991). “Determi-nants of Portfolio Performance II: An Update,” FinancialAnalysts Journal 47, 40–48.

Brown, K., Garlappi, L., and Tiu, C. (2009). “AssetAllocation and Portfolio Performance: Evidence from Uni-versity Endowment Funds,” Journal of Financial Markets,forthcoming.

Brown, K., Harlow, V., and Starks, L. (1996). “Of Tour-naments and Temptations: An Analysis of ManagerialIncentives in the Mutual Fund Industry,” Journal of Finance51, 85–110.

Brown, S., Fraser, T., and Liang, B. (2008). “Hedge Fund DueDiligence: A Source of Alpha in a Hedge Fund PortfolioStrategy,” Journal of Investment Management 6, 23–33.

Chen, J., Hong, H., Huang, M., and Kubik, J. (2004). “DoesFund Size Erode Mutual Fund Performance? The Role ofLiquidity and Organization,” American Economic Review94, 1276–1302.

Chow, G. and Kritzman, M. (2001). “Risk Budgets,” Journalof Portfolio Management 27, 56–60.

Clarke, R., de Silva, H., and Thorley, S. (2006). “The Fun-damental Law of Active Portfolio Management,” Journal ofInvestment Management 4, 54–72.

Clarke, R., de Silva, H., and Wander, B. (2002). “RiskAllocation Versus Asset Allocation,” Journal of PortfolioManagement 29, 9–30.

Grinold, R. (1989). “The Fundamental Law of ActiveManagement,” Journal of Portfolio Management 15, 30–37.

Grinold, R. and Kahn, R. (2000). Active Portfolio Manage-ment, 2nd edn. New York: McGraw-Hill.

Griswold, J. (ed.) (2008). Commonfund Benchmarks Study:Educational Endowment Report, Wilton CT: CommonfundInstitute.

Harlow, V. (1991). “Asset Allocation in a Downside-RiskFramework,” Financial Analysts Journal 47, 28–40.

Harlow, V. and Brown, K. (2006). “The Right Answer tothe Wrong Question: Identifying Superior Active PortfolioManagement,” Journal of Investment Management 4, 15–40.

Ibbotson, R. and Kaplan, P. (2000). “Does Asset AllocationPolicy Explain 40, 90, or 100 Percent of Performance?”Financial Analysts Journal 56, 26–33,.

Ineichen, A. (2007). Asymmetric Returns: The Future of ActiveManagement, New York: John Wiley & Sons.

Kane, A., Kim, T., and White, H. (2003). “Active Portfo-lio Management: The Power of the Treynor-black Model,”Working Paper, University of California, San Diego.

Leibowitz, M. (2005). “Alpha Hunters and Beta Grazers,”Financial Analysts Journal 61, 32–39.

Lo, A. (2008). “Where do Alphas Come From?: A Measureof the Value of Active Investment Management,” Journal ofInvestment Management 6, 6–34.

Pearson, N. (2002). Risk Budgeting: Portfolio Problem Solvingwith Value-at-Risk, New York: John Wiley & Sons.

Qian, E. and Hua, R. (2004). “Active Risk and InformationRatio,” Journal of Investment Management 3, 20–34.

Ross, S. (1976). “The Arbitrage Theory of Capital AssetPricing,” Journal of Economic Theory 13, 341–360.

Sharpe, W. (1964). “Capital Asset Prices: A Theory of MarketEquilibrium Under Conditions of Risk,” Journal of Finance19, 425–442.

Swensen, D. (2009). Pioneering Portfolio Management: AnUnconventional Approach to Institutional Investment, 2ndedition, New York: Free Press.

Treynor, J. and Black, F. (1973) “How to Use Security Anal-ysis to Improve Portfolio Selection,” Journal of Business 46,66–86.

Walter, I. (2004). “Conflicts of Interest and Market DisciplineAmong Financial Services Firms,” European ManagementJournal 22, 361–376.

Waring, M. and Siegel, L. (2003). “The Dimensions of ActiveRisk,” Journal of Portfolio Management 29, 35–51.

Winkelman, K. (2003). “Issues in Strategic Asset Allocation,”In Litterman, R. (ed.), Modern Investment Management: AnEquilibrium Approach, Hoboken, NJ: John Wiley & Sons,2003.

Keywords: Risk budgeting, active–passive alloca-tions, investment performance, endowment funds.


journal of investment management joim · a risk management question, ... after showing how...

Documents