To augment the literature, the author explores how characteristics and norms of decision-making committees, typically consisting of volunteers, affect portfolio diversification and risk-adjusted performance of U.S. university endowment portfolios. Specifically, this study examines how the diversity of investment expertise among committee members, in conjunction with norms of a learning commitment and open-mindedness, contribute to a decision-making process that influences portfolio diversification and performance. The selection of these particular variables was informed by studies outside of the investment industry regarding group decision-making (Janis 1972
; Amason 1996
; Forbes and Milliken 1999
), and the author’s interviews of endowment representatives on campuses (Lord 2014a
Markowitz’s Modern Portfolio Theory is proposed as the decision-making framework, measured by a scale developed by the author and endorsed by Mr. Markowitz. The committee’s use of the theory is proposed as an intermediary of the effect of committee characteristics and norms on portfolio diversification. Diversification, in turn, is shown to have a direct effect on risk-adjusted performance.
This study focuses on the investment committee since it is the body that establishes investment goals and typically sets the asset allocation policy for the endowment. Research within the university endowment domain has found that investment committee members’ experience on other boards and their business background are correlated with the degree of diversification of the endowment portfolio (Brown et al. 2011
). The current study, however, is more specific about the committee members’ background, focusing on the degree to which members have expertise in a variety of asset classes, such as domestic and international equities, fixed income, and a range of alternative investments.
Annual studies of college and university endowments by the National Association of College and University Business Officers (NACUBO 1990–2008
) demonstrate a positive correlation between size and performance. Large endowments, it is argued, have greater resources to hire talented investment professionals and greater access to top managers, particularly in alternative asset classes (Lerner et al. 2008
). Those researchers also found that performance is highly correlated with the academic quality of the student body, as measured by SAT scores. SAT scores may serve as a proxy for the overall skill of the university’s administration, the wealth and connections of the alumni, and the prestige of the university brand. Since committee members are often drawn from alumni, their academic abilities may contribute to better decision-making processes and outcomes (Lerner et al. 2008
The current study, while recognizing the value of factors mentioned above, differs from other university endowment research in its focus on the inner-workings of the committee, and particularly with regards to group norms that affect how members value the importance of continual learning and how they use open-minded dialogue and Portfolio Theory principles to help make asset allocation decisions.
Outside of the university endowment arena, a broad study of mutual fund performance (Bär et al. 2011
) found that team-managed funds, as contrasted with single-manager funds, had more diversified portfolios and moderate performance. Single-manager funds tended to have more concentrated portfolios and “extreme” performance (both superior and inferior) than team-managed funds. In contrast, none of the university endowments in the current study has a single manager. Moreover, team-led managers of mutual funds typically have different goals and constraints than university endowment committees, thus influencing their allocations differently.
Understanding factors related to portfolio performance is critical for college and university leaders since the endowment’s returns can profoundly affect the amount of student financial aid, administrative staff size, facilities maintenance, programming, and mix of tenure-track and adjunct faculty (Brown et al. 2010
). Investment performance of university endowments has taken on added significance due to recession-induced declines in financial markets, donor giving, and state allocations to higher education. The author employs risk-adjusted returns as the dependent variable, rather than absolute returns, in order to make performance comparisons irrespective of differing goals and constraints.
Modern Portfolio Theory, referred to hereafter as simply as Portfolio Theory, is based on mean-variance analysis (Markowitz 1952
), and provides a framework for constructing portfolios that balances the twin goals of performance and risk management. Specifically, the theory posits that diversification promotes efficient portfolios that either: (a) provide greater returns given a specified risk level; or (b) reduce risk for a specified level of return. A number of software programs have been developed to help investors optimize the risk-return profile of their portfolios, with recommended asset allocations given their goals and constraints.
University endowment portfolios have become increasingly diversified in recent years across geographic regions and asset classes, as shown in Table 1
As recently as 1993, the average U.S. endowment participating in the annual survey by NACUBO
) had allocated 87% of its portfolio to traditional asset classes of fixed income and publicly-traded equities (U.S. and international), 9% to cash and “other,” and only 4% to alternative investments defined as hedge funds, private equity, real estate, venture capital and natural resources. By fiscal year-end 2008, the average allocation to alternative investments had grown to 23.7% with hedge funds claiming over half of the alternative allocation. Other notable changes over the 15 years are larger allocations to international equities (with smaller allocations to U.S. equities) and smaller allocations to fixed income.
Asset allocations differ remarkably by the size of the portfolio: in 2008, the largest endowments (those with portfolios greater than $
1 billion) had allocated 48% to alternative investments, whereas the smallest endowments (less than $
25 million) had allocated just 7% to those investments. Researchers caution against naive imitation of large endowments’ asset allocations because alternative assets vary considerably in their performance, and expertise is required to make good decisions (Lerner et al. 2008
). Investors also need to be aware of illiquidity risks associated with private investments (Brown et al. 2010
). Leibowitz et al.
) have demonstrated that the benefits of endowments’ allocations to alternative investments are derived predominantly from the additional return expectations rather than from reduced volatility, because the addition of alternative investments has little effect on long-term portfolio volatility.
For decades, investment researchers have attempted to ascertain the relative determinants of performance coming from asset allocation policy, active asset allocation (often called market timing) and selection of specific investments/securities within the asset classes (Brinson et al. 1986
; Brinson et al. 1991
; Ibbotson and Kaplan 2000
; Kritzman and Page 2003
). This study, in contrast, examines only on the asset allocation component, and, particularly, the degree to which the portfolio is broadly diversified across a variety of asset classes. While seeking to confirm that broadly diversified portfolios provide higher risk-adjusted returns than do less diversified portfolios, the greater focus in this study is on characteristics and behaviors of the committee in making decisions for more broadly diversified portfolios. The study does not attempt to assess the portions of return resulting from market timing and individual investment/security selection.
In a novel approach, we focus on factors that contribute to greater diversification and stronger risk-adjusted performance among endowments of similar size. This approach provides improved isolation of the factors that are hypothesized to have positive effects on diversification and performance, irrespective of the portfolio’s size. Other endowment research has established links between performance and access to top managers, “Ivy League” status, greater alumni giving, higher student-body SAT scores, and the ratio of endowment dollars per student (Lerner et al. 2008
). Yet those findings provide few actionable steps for the endowment committee in attempting to improve risk-adjusted performance. The research question posed in this study is a practical one that has not been examined empirically: How can colleges/universities foster an environment that leads to broad diversification and higher risk-adjusted returns of their endowment portfolios? The author argues that selecting investment committee members with expertise in diverse asset classes, promoting an open-minded and learning environment, and employing the principles of Portfolio Theory can play key roles in reaching those objectives, as conceptualized in Figure 1
3. Research Methods
The sample was drawn from 650 colleges and universities that have participated regularly in industry surveys about endowment investment practices and for which a minimum of five-year (2004–2009) performance data were available. A total of 191 colleges/universities responded to the survey for a response rate of 29%. That number was reduced to 179, or 27.5%, after eliminating nine cases due to incomplete surveys and three outliers. Sample size of 179 is considered adequate as it comfortably exceeds the minimum recommendation of five observations relative to the number of independent variables (Hair et al. 2010
Characteristics of respondents, shown in Table 2
, revealed that all but four respondents were finance, foundation, or investment officers; two were outsourced Chief Investment Officers and two were investment committee members. On average, respondents have served 11 years in an endowment-related role with the college/university. Respondents were from both public (39%) and private (61%) institutions and the size of endowments spanned all of the categories in the annual NACUBO study, from less than $
25 million to greater than $
1 billion. The average endowment size of survey participants as of fiscal year-end 2009 was $
332 million, compared to $
306 million in the 2009 NACUBO-Commonfund Study of Endowments (NACUBO 2009
The investment committees in the sample have played important roles in key decisions concerning the management of the portfolio: 68% of respondents indicated that the committee made final decisions about hiring/firing managers and 67% said the committee made the final decision about hiring/firing consultants. Sixty-two percent of respondents indicated that the investment committee made the final decisions regarding policy asset allocations over the five-year measurement period, followed by the governing board of trustees/foundation, 34%. In response to the question, “Which body had the greatest impact on performance over the 2004–2009 period”, responses were distributed more widely: investment committee, 41%; consultant, 25%; staff (such as Chief Investment Officer or Chief Financial Officer), 16%; outsourced CIO or Investment Management Firm, 13%; board of trustees/foundation, 3%; and other, 2%.
To determine if the sample was representative of all 650 colleges with five-year performance data in the 2009 NACUBO-Commonfund survey, we conducted an independent samples t-test of the means of the five-year annualized returns. No significant difference was observed between the means (t = 0.44; df = 826; p > 0.05). The mean return from the NACUBO-Commonfund study was 2.7%, s = 2.8%, while the mean of this sample was 2.6%; s = 2.1%.
3.2. Data Collection and Screening
Measures for content validity were assessed using expert opinions of 12 asset management professionals. For internal validity, the author used a pretest to assess the nature of the relationships among the constructs and modified problematic items as necessary. This strategy helped ensure that the measurement scales possessed adequate content validity, while permitting the testing of their psychometric properties (i.e., scale reliability and construct validity).
Empirical data to test the hypothesized relationships were obtained by using an electronic survey. Emails soliciting participation were sent to 650 colleges and universities, all of which had participated in the 2009 NACUBO-Commonfund Endowment Study and/or in previous annual surveys sponsored solely by NACUBO and for which five-year performance data through mid-2009 were available. Emails were addressed to financial officers requesting survey participation by a “key informant:” someone who has regularly attended committee meetings for at least several years and is very knowledgeable about the committee’s responsibilities and decision-making practices, as well as the investment experience of committee members. The solicitation email suggested that either the university financial officer most involved with the endowment or the investment committee chair would be an ideal respondent. Three follow-up emails were sent in intervals of approximately 12 days during September and October 2010.
After eliminating incomplete surveys and outliers, missing data on remaining cases were less than 1% and were replaced with median values for the variable. Although the data were slightly skewed, there were no serious violations of normality assumptions detected. Correlations between factors were all less than 0.60 and all variance inflation factors were below 4.0, indicating a lack of problems with multicollinearity. Homogeneity of variance was tested using Levene’s test. As all IVs and mediator were significant, we cannot reject the null hypothesis that the variance of the groups is equal.
The questionnaire consisted of 21 items used in scales for the constructs, provided in Appendix A
. In addition, respondents were asked to provide their endowment’s five-year annualized total return, using the same number they provided in the 2009 NACUBO-Commonfund study. Survey participants were instructed to base their responses over the period covering mid-2004 to mid-2009. That timeframe was chosen since: (1) it was a period of time for which the author gratefully had full access to NACUBO studies (certain data specific to individual universities are not available to the general public); (2) it contained different market environments with periods of both positive and negative endowment performance; (3) it covered a span in which the majority of the author’s survey recipients were expected to have been involved with the endowment; (4) it permitted a sufficient number of respondents with five-year performance data in order to conduct meaningful analysis.
3.4. Independent Variables
The scale items for the three independent variables (diverse expertise, learning commitment and open-mindedness) employed a 7-point Likert scale ranging from Very Strongly Disagree to Very Strongly Agree. Both learning commitment and open-mindedness were adapted from existing scales (Sinkula et al. 1997
; Calantone et al. 2002
). Among the learning commitment items were: “The committee believed learning is a key commodity for long-term success” and “The committee was committed to learning about successful endowment practices.” Open-mindedness items included: “Committee members frequently questioned their biases about investing” and “The committee was not afraid to reflect critically on investment-related assumptions it made.” Items for diverse committee expertise were informed by extensive interviews from an earlier qualitative study (Lord 2014a
), theory (Forbes and Milliken 1999
), and pre-testing with investment professionals. Committee expertise items sought to identify whether members had expertise across a broad variety of investments including traditional and alternative asset classes, as well as public and private.
3.5. Portfolio Theory
Portfolio Theory is hypothesized as an intermediary variable serving as a decision-making process for portfolio allocations. The development of the Portfolio Theory measure was informed by theory (Markowitz 1952
) and by pretesting with 12 investment professionals. Items inquired about members’ consideration of investments’ expected returns, variance of returns and correlations of returns when making portfolio decisions. Additional questions included variations of those items, such as: “The committee considered the risk-return characteristics of the overall portfolio when making changes in its composition” and “The committee carefully considered the downside risks of existing and potential investments”.
3.6. Diversification of Portfolio
Since portfolio diversification and size of portfolio have been linked in previous studies (NACUBO 1990–2008
), the measure of portfolio diversification in this study controlled for size by asking for participants’ perception of their endowment’s degree of diversification relative
to the average diversification of peers in their size category in annual industry studies. Size categories in the NACUBO studies are as follows: (1) <$
25 mil; (2) $
25 mil to $
50 mil; (3) $
50 mil to $
100 mil; (4) $
100 mil to $
500 mil; (5) $
500 mil to $
1 bil; and (6) >$
Supporting evidence of participants’ awareness of other endowments’ degree of diversification is based on an 86% affirmative answer to the following survey question: “Has the committee typically compared your institution with others in the NACUBO and/or NACUBO-Commonfund study over the past several years?” Of the 179 participants in this study, 154 responded “yes”, 19 responded “no”, and six did not respond.
Three items were used to measure portfolio diversification relative to size peers over the recent 10-year, 5-year, and 1-year periods; response choices ranged from 1-significantly less diversified to 5-significantly more diversified. Although primarily interested in the 5-year period, the author included both a longer and shorter period for a fuller perspective; industry surveys have documented the steady, incremental march toward greater portfolio diversification, both on a total industry level and on the category size level (NACUBO 1990–2008
). It is thus logical to assume that participants’ perceptions of their endowments’ diversification relative to their size peers would be largely consistent over the time periods.
Note that in a related study (Lord 2014b
), the author used actual asset allocation data from the 2008 NACUBO study to assess the degree of diversification across asset classes. The significant degree of correlation between that data and the responses in the current study helps to support the validity of using respondents’ perceptions
of their endowments’ relative degree of diversification.
For performance measurement, the survey requested annualized total returns of endowments over 1, 3, 5, and 10-year periods ended fiscal 2009; the 5-year data used in this study were crossed-checked with data in the 2009 NACUBO-Commonfund Endowment Survey. In order to obtain risk-adjusted performance the author divided the five-year annualized return of each endowment by the standard deviation of the annual returns over the five years. Since absolute endowment size has already been linked to performance in industry studies (NACUBO 1990–2008
), we chose to measure risk-adjusted performance of each university relative
to the risk-adjusted average performance of peers in their own size category. Therefore, both portfolio diversification and risk-adjusted performance are relative measures.
3.8. Control Variables
The number of endowment staff with professional investment expertise as well as the absolute endowment size were controlled for since these variables have been linked previously to investment performance (NACUBO 1990–2008
3.9. Exploratory Analysis
An Exploratory Factor Analysis was conducted simultaneously with all items for the latent factors using Principal Axis Factoring with Promax rotation. The Kaiser-Meyer-Olkin measure of 0.904 indicated sampling adequacy for factor analysis. Analysis of initial communalities revealed that all items exceeded the threshold of 0.50. An unconstrained five-factor model emerged from the extraction process. All five factors had eigenvalues greater than 1.0 and explained 81% of total variance. The pattern matrix, as shown in Appendix B
, revealed that all items loaded highly on their intended factors with minimal cross-loading. Appendix C
contains Descriptive Statistics and Appendix D
contains the correlation of measures for constructs.
3.10. Confirmatory Factor Analysis
Confirmatory Factor Analysis (CFA) was conducted next to validate the initial factors’ structure and to assess model fit, which was deemed acceptable (cmin/df of 1.427, GFI of 0.892, CFI of 0.977, PCFI of 0.819 and RMSEA of 0.049 with PCLOSE of 0.537). Three pairs of items were covaried, each within its hypothesized construct, contributing positively to model fit. A chi-square difference test confirmed measurement invariance of two groups (top performers and bottom performers). The CFA diagram is provided in Appendix E
. As seen in Table 3
, loadings ranged from 0.66 to 0.99 and all were statistically significant at p
3.11. Common Method Variance
The first test for common method bias used a post hoc procedure suggested by Podsakoff and Organ
). Using the principal components factor analysis, evidence of common method bias exists when a single factor emerges from the analysis, or one general factor accounts for the majority of the covariance in the interdependent and dependent variables. Those conditions do not exist in this study’s factor analysis. A second test for CMB involved the addition of a marker variable (Lindell and Whitney 2001
), which reduced common method bias to 7.3%.
3.12. Reliability and Validity Analysis
To assess the measurement model’s reliability, discriminate and convergent validity, the author used the composite reliability coefficient (CR), average variance extracted (AVE) and the shared variance (Maximum Shared Variance, MSV, and Average Shared Variance, ASV). Table 4
indicates that the model met the threshold for item reliability (CR > 0.70) as well as those for convergent and discriminant validity, both before and after controlling for common method variance (CMV). To estimate the common method variance the author used the correlation between the marker variable and the other factors to recompute the reliability and validity measures shown in Table 4
. All of the measurement model’s constructs retained their reliability, convergent validity, and discriminate validity even after controlling for common method variance. Internal consistency was confirmed with Cronbach’s alpha coefficients all well above 0.7. The measures support the reliability as well as the convergent and discriminant validity of the constructs.
3.13. Hypotheses Testing
Hypotheses were tested using covariance-based structural equation modeling (SEM). SEM is statistical software tool that can simultaneously evaluate multiple variables and their relationships. It can be used for concept and theory development as well as for hypothesis testing (Hair et al. 2014
). The SEM software used for this study is IBM SPSS AMOS (analysis of a moment structure). Results for the full sample are shown in Figure 2
(before trimming insignificant paths) and also in Table 5
, both before and after trimming insignificant paths. The trimmed model has excellent fit with CMIN/DF of 1.11, CFI of 0.996, and RMSEA of 0.025.
4. Results and Discussion
Hypothesis 1. This hypothesis is supported. This finding has sound logic in that one would expect committee members with expertise in particular asset classes to generally want some representation in the portfolio of those asset classes. This would be particularly the case when the endowment follows a policy asset allocation with a long-term horizon. In most instances, the policy asset allocation permits a range of weights in the portfolio for each asset class, and so the committee would have some leeway, if desired, to alter the weight somewhat based on factors such as valuation and economic conditions.
Hypothesis 2. This hypothesis is not supported. Perhaps in many situations, each committee member has expertise in only one asset class and is not accustomed to considering expected returns, variance of returns, and correlations of returns involving the other asset classes.
Hypothesis 3. This hypothesis is supported. Committee members with a commitment to learning, which includes a belief that learning is a key component to success, will devote considerable time and effort to learn about and improve their understanding of principles of portfolio construction. In this model, perhaps the hardest “work” is making estimates about expected returns of the various asset classes, variance of returns, and correlation of returns. Moreover, the estimates need to be revisited at times due to changing valuations and economic conditions, changing weights of each asset class in the portfolio, etc. In a sense, the “commitment to learning” variable also represents a level of diligence and dedication, or, quite simply, a willingness to do the work.
Hypothesis 4. This hypothesis is not supported. There is a modest positive correlation between the two factors, but it is not significant.
Hypothesis 5. This hypothesis is supported. The open-mindedness construct represents a willingness for committee members to continually question their biases and assumptions. It encourages critical reflection and a willingness to express differing opinions. Since the use of Portfolio Theory principles requires making important estimates, it seems reasonable that having an open-minded committee would hash out their views in making these determinations. Note that while Open-Mindedness was not hypothesized to have a direct effect on portfolio diversification, it does, at a minimum, have an indirect effect via Portfolio Theory.
Hypothesis 6. This hypothesis is supported. While the Portfolio Theory component of estimating returns might influence committee decisions about the weights to assign to each asset class, the other two components (variation of returns and correlation of returns) likely play a bigger role in affecting the diversification of the portfolio. For example, asset classes with sizable return variations might need to be limited to a certain level to avoid making the portfolio riskier than the university desired. Also, asset classes that have significant return correlations may need to be limited to avoid the risks of over-concentration in the behavior characteristics of those asset classes. Those are just two examples of situations where use of Portfolio Theory would result in broader portfolio diversification.
This hypothesis is supported in the final results after trimming the weakest path in the hypothesized model. Considerable research has documented the significant and positive correlation between portfolio diversification and performance, and the significant positive relationship between size and each of those. However, this finding differs from previous ones in that it examines the relationship within each category of portfolio size. Thus, even among portfolios of similar size, broader diversification leads to higher risk-adjusted returns. Note in the top portion of Table 5
that Portfolio Theory does NOT have a significant effect on portfolio diversification before trimming insignificant paths. However, the SEM analysis requires trimming (eliminating) insignificant paths one at a time, starting with the weakest path. The weakest path was that from Learning Commitment to Diversified Portfolio, and once that path was trimmed, the path from Portfolio Theory to Diversified Portfolio became significant.
The control variable for the number of full-time professional investment staff does have a significant effect on the relative performance dependent variable. However, the second control variable—absolute endowment size—does not have a significant effect on the performance of endowments relative to their size category. The author assumes that the performance variable, which measures performance relative to other portfolios in the same size category, already provides a sufficient control for size.
Despite its contributions, the study has some notable limitations providing opportunities for future research. First, the measures of committee characteristics and practices were based on key informant assessments that may be biased. Next, the study does not include certain components of return that are well recognized, specifically active (or tactical) asset allocation and investment/security selection. The other key component of return, called policy asset allocation, is roughly represented in the study by the variable “Diversified Portfolio.” Policy asset allocation establishes long-term allocations within ranges for each asset class. Active asset allocation and investment/stock selection were not included in this study due to inaccessibility to data and complexity.
In addition, numerous other factors that likely contribute to performance—such as criteria and processes for selecting consultants and investment managers—were not included in the study. As indicated earlier, for many endowments the influence of consultants was considered a key factor for performance and it would be useful to gain a better understanding of the nature and effect of that influence. Another important limitation of this study is that it did not analyze the influences of university governing boards other than noting that approximately one-third of respondents indicated that the governing board controls the final decision over asset allocation policy. Further research on the nature of governing board’s influence on portfolios appears to be warranted.
Other variables—such as academic quality of students and level of alumni donations—that have been found in other studies to correlate to performance were not controlled for in this study. Those factors may contribute positively to the expertise sought in selecting not only committee members also external consultants and investment managers.
Another limitation is that the study does not account for the fact that different universities may have different goals and objectives for the portfolio. For example, certain endowments are used to support student scholarships and the portfolios need to produce fairly stable performance with above-average income. Other endowments of universities with plans for long-term growth may need to be invested more aggressively. Therefore, while the dependent variable of risk-adjusted returns (relative to other endowments of similar size), is convenient for making comparisons, it likely does not represent the primary objective for many endowments.
Lastly, the direction of relationships cannot be certain. For example, does “greater expertise across a variety of asset classes” lead to “greater portfolio diversification” or does the relationship occur in reverse order? It would seem likely that in the early stages of developing more diversified portfolios, the direction specified in the model would facilitate the understanding of various asset classes in order for the committee to make decisions about adding more asset classes. It is important to note that professional, external consultants also play a key role, not examined in the current study, in explaining various asset classes and their roles in a portfolio. Later on, as the portfolio becomes more diversified and complex, and members rotate off the committee, the selection of new members may include a consideration that they already have sufficient expertise to understand the existing portfolio and, hopefully, to make improvements.