Financial Analysts Journal

ISSN: 0015-198X (Print) 1938-3312 (Online) Journal homepage: https://www.tandfonline.com/loi/ufaj20

A Sensible Mutual Fund Selection Model

Hakan Saraoglu & Miranda Lam Detzler

To cite this article: Hakan Saraoglu & Miranda Lam Detzler (2002) A Sensible Mutual Fund
Selection Model, Financial Analysts Journal, 58:3, 60-72, DOI: 10.2469/faj.v58.n3.2538

To link to this article: https://doi.org/10.2469/faj.v58.n3.2538

Published online: 02 Jan 2019.

Submit your article to this journal

Article views: 1

Citing articles: 2 View citing articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=ufaj20
 A Sensible Mutual Fund Selection Model
Hakan Saraoglu and Miranda Lam Detzler
We present a rigorous framework for asset allocation and selecting mutual
funds that takes into account the unique preferences and constraints of an
individual investor. The framework is based on the analytic hierarchy
process (AHP), and the model generates reasonable asset-allocation
recommendations and identifies the most suitable funds within an asset
class. We include sample mutual fund selection for a hypothetical investor.
A mutual fund selection model that uses the AHP framework is flexible, is
user friendly, and ensures consistency throughout the portfolio decision
process.

odern portfolio theory (MPT) states investor in determining an appropriate asset allo-

M that the investment decision process

can be separated into two indepen-
dent processes. In one, investment
professionals, such as mutual fund managers, spe-
cation and selecting mutual funds for a complete
portfolio.
Mutual funds provide diversification, divisi-
bility, low transaction costs, record keeping, and
cialize in constructing a variety of risky portfolios. professional management for the individual inves-
In a second process, individual investors choose tor. These features have helped propel the popular-
complete portfolios by combining the optimal risky ity of mutual funds to historic heights in the past
portfolio, based on their risk tolerances, and the decade. As of March 2000, more than 7,800 mutual
risk-free asset. Finance research has traditionally funds were operating in the United States, with
focused on the first part of the investment process, combined total assets under management of $7.294
especially on identifying some risky portfolio that trillion (see Mutual Fund Fact Book 2000). For com-
would be universally optimal for all investors. parison, approximately 8,600 stocks are traded on
In the real world, however, the investment the NYSE, Amex, and Nasdaq.
process is more complicated; in addition to risk and The proliferation of mutual funds has made
return, factors such as taxes and human capital choosing the right funds a challenge to many inves-
must be considered. Because different investors
tors. In response, many magazines and news ser-
face different constraints in their investment deci-
vices designed to assist investors in mutual fund
sions, a risky portfolio that is universally optimal
selection have emerged. These sources provide per-
may be impossible to find. Investors are left to their
formance statistics and fund attributes, such as
own judgments when determining the proper risk
information on fund managers, expense ratios, and
level for themselves and constructing their own
turnover, but an individual investor seldom has the
complete portfolios. But many individual investors
time or the expertise to analyze the vast amount of
may not be competent at these tasks. According to
data available on mutual funds. Therefore, some
a 1994 survey by Merrill Lynch & Company, more
financial software and finance Web sites now offer
than 80 percent of the respondents failed a test of
basic finance knowledge (see Merrill Lynch Finan- mutual fund screening tools. Unfortunately, these
cial Literacy Index Analysis 1994). For such reasons, tools range from the inadequate to the useless. First,
Brennan (1995) argued that examining the invest- few investors have sufficient financial knowledge
ment decision process from the individual inves- to input the appropriate values for the screening
tor’s perspective is a much needed area of research. variables. Second, the tools do not consider the
We propose a model that assists an individual preferences of individual investors; the tools
implicitly assume that each screening variable is
equally important to all investors. Finally, such a
large number of funds may meet the screening
Hakan Saraoglu is assistant professor of finance at Bry-
ant College, Smithfield, Rhode Island. Miranda Lam criteria (a common occurrence) that investors still
Detzler, CFA, is assistant professor of finance at the need to make selections from the list.
University of Massachusetts, Boston. An alternative to these tools is the assistance of
a professional financial advisor or a full-service

60 ©2002, AIMR®
 A Sensible Mutual Fund Selection Model

broker who can provide customized advice. The academic research, and historical mutual fund
financial planning process typically begins with the data, the result is a user-friendly yet rigorous
investor filling out a questionnaire to pinpoint the model. We believe that the AHP mutual fund selec-
investor’s risk aversion, investment horizon, tion model described here will be useful to individ-
investment experience, tax status, and financial sta- ual investors and to the financial advisors who
tus. Based on this information, the financial advisor make mutual fund recommendations.
recommends an asset-allocation strategy and sug- An important distinction between the AHP
gests specific mutual funds (or stocks and bonds) and the MPT approaches is that in the AHP, the
for each asset class. A problem with this approach focus is on the preferences of the investor, whereas
is that financial advisors often have high potential in the MPT approach, the focus is on only two
liability because disgruntled investors can sue for dimensions of an individual’s preferences—risk
damages if they believe they received unsuitable and return. In addition, unconstrained mean–
investment advice (Bolster, Janjigian, and Trahan variance optimization often generates portfolios
1995). Demonstrating that an investment recom- with extreme weights and is very sensitive to esti-
mendation is suitable can be difficult because argu- mation errors. For a detailed discussion of the dif-
ments as to suitability are largely subjective. One ferences between the AHP approach and the MPT
solution is to make the financial planning process approach, see Bolster and Warrick (1999).
more objective.
We propose and illustrate a structural approach Analytic Hierarchy Process
to selecting mutual funds that is based on the ana-
The AHP (Saaty 1980) helps decision makers sys-
lytic hierarchy process (AHP) and overcomes some
tematically structure a complex multiattribute
of the shortcomings of typical screening tools. Our
problem. It has been applied to many disciplines,
approach thus adds to financial advisors’ toolkit an
such as microcomputer selection (Arbel and
objective procedure for choosing mutual funds. The
Seidmann 1984), budget allocation (Sinuany-Stern
example we develop demonstrates how to adapt the
1984), and project selection (Johnson and Hihn
AHP to solve the mutual fund selection problem.
1980). A variety of financial problems have also
The AHP model has the following distinct been solved using the AHP methodology, includ-
advantages. Perhaps the most important contribu- ing assignment of sovereign debt ratings (Johnson,
tion of the model is that it provides a systematic Srinivasan, and Bolster 1990), determination of
approach to ranking mutual funds for individuals investor suitability (Bolster, Janjigian, and Trahan),
based on each individual’s unique investment selection of a life insurance contract (Puelz 1991),
objectives and constraints. The complete portfolio and determination of an optimal portfolio mix
of funds selected by the AHP model is customized (Khaksari, Kamath, and Grieves 1989).
for a particular investor. Second, the AHP prevents The AHP methodology consists of the follow-
the investor from making inconsistent preference ing four major steps:
assignments. Because fund selection involves more 1. Develop the hierarchical structure:
than one parameter (for example, it may involve
• Mission.
investment objectives, tax efficiency, risk, and
expense ratios), enforcing consistency in the deci- • Selection criteria.
sions is difficult if the selection method is unstruc- • Alternatives.
tured.1 Olsen (1998, p. 16) noted that “when 2. Assign a relative importance of each selec-
decisions become complex, decision makers shift tion criterion to the mission.
toward using rules of thumb and noncompensa- 3. Rank alternatives under each criterion.
tory procedures (that is, decision rules in which 4. Rank each alternative’s contribution to the
trade-offs do not involve weighting of attributes).” mission.
The AHP helps steer investors away from rules of Complex problems will entail several hierar-
thumb that do not reflect their personal prefer- chies, each with its mission, selection criteria, and
ences. Unlike the typical screening tool, which pro- alternatives. For example, the mutual fund selec-
duces a list of funds that meet certain criteria, the tion problem can be divided into two hierarchies:
AHP ranks the selected mutual funds on the basis The first hierarchy addresses alternatives at the
of the investor’s preferences, making the final deci- asset-class level (asset allocation), and the second
sion much easier. Third, this approach minimizes hierarchy addresses alternatives at the fund level
the amount of technical input required from inves- (fund selection). The mission of the overall exercise
tors. Because the model can integrate recommen- is to identify mutual funds that are suitable for a
dations by a financial advisor, findings from particular investor.

May/June 2002 61
Financial Analysts Journal

Mutual Fund Selection Figure 1 presents the two hierarchies in the mutual
fund selection problem. Each hierarchy consists of
How to choose the right mutual funds has been the
three levels—the mission of the hierarchy (given at
subject of numerous books and countless articles.2
the top), the selection criteria, and the available
Some are “how-to” guides that provide rules of
alternatives. In the first hierarchy, the mission is to
thumb based on the authors’ intuition and experi-
obtain a suitable asset-allocation recommendation
ence combined with findings from academic
for an investor. The selection criteria are the inves-
research, and others claim to have discovered strat-
tor’s investment objectives and constraints, and the
egies for identifying “superior” funds. An impor-
alternatives are the asset classes. In the second hier-
tant distinction is the difference between choosing
archy, the mission is to obtain the most suitable
the right funds for an investor and predicting mutual fund within each asset class; Figure 1 uses
which funds will have the best performance. Previ- as an example the “growth and income” class. The
ous studies have shown that mutual fund manag- selection criteria are based on the structural and
ers cannot outperform unmanaged indexes of U.S. operational characteristics of the funds, and the
stocks, U.S. bonds, international stocks, or interna- alternatives are the funds available in each asset
tional bonds (Malkiel 1995; Blake, Elton, and Gru- class. The hierarchical structure in Figure 1 is
ber 1993; Eun, Kolodny, and Resnick 1991; and intended to be a simplified model for demonstrat-
Detzler 1999) and that superior performance sel- ing the AHP framework; it can be easily modified
dom persists (Elton, Gruber, and Blake 1996; Grin- to include additional objectives and constraints.4
blatt and Titman 1992; Carhart 1997). The purpose Exhibit 1 summarizes the process of solving the
of this study is to assist investors in their asset- problem represented by each hierarchy.
allocation decisions using mutual funds as invest-
ment vehicles, not to predict future performance. Investment Objectives. The decision pro-
A major deficiency of existing fund-screening cess for mutual fund selection starts with the inves-
models and how-to books on mutual funds is that tor’s investment objectives and constraints. We
they do not incorporate an investor’s preferences. consulted numerous sources, including textbooks,
For example, Bogle on Mutual Funds, a national magazines, finance Web sites, SEC rules, and gov-
bestseller, provides a list of services to be consid- ernment publications, on the most common invest-
ered when choosing a fund company but states, “In ment objectives. Although the advice from each
the end, it is up to you (the investor) to determine source differs slightly, these sources generally
the relative importance of the many contributing agree that investment objectives should be related
factors” (Bogle 1994, p. 60.) To carry out such a to the investor’s risk aversion, investment horizon,
determination, a mutual fund selection model lifestyle, and demographic profile. The sample
based on the AHP allows an investor to specify the model in Figure 1 considers six investment objec-
relative importance of each factor in a systematic tives—capital appreciation in 5 years, capital
manner. Such a systematic approach is critical appreciation in 15 years, current income, preserva-
because of the large number of factors that must be tion of principal, international diversification, and
considered when selecting mutual funds. For tax advantage.
example, the U.S. SEC requires a fund company to The investor assesses the relative importance of
provide data to potential investors about 11 items.3 the investment objectives by using a pairwise com-
A 1997 Investment Company Institute survey (see parison scale. Exhibit 2 presents the scale typically
Mutual Fund Fact Book 1997) found that an average used in the AHP. With six investment objectives, the
mutual fund investor considers 13 items of fund- investor needs to make 15 pairwise comparisons. A
related information before investing. But studies sample questionnaire and responses from a hypo-
have shown that when individuals have to consider thetical investor are in Appendix A. Panel A of
more than seven variables in a decision, they may Table 1 presents the preferences of the hypothetical
make inconsistent preference assignments (Miller investor in a matrix format. For example, compar-
1956). The AHP enforces consistency in assigning ing current income with capital appreciation over
preferences by asking the investor to make only five years (first column, third row of the matrix), the
pairwise comparisons, thus avoiding the need for investor assigned a preference of 4, indicating that
the investor to consider a large number of factors current income is more important.
simultaneously. Before proceeding to the next level of the hier-
archy, the investor or financial advisor calculates
The AHP-Based Mutual Fund Selection the consistency of the pairwise comparisons by
Model. The first step in the AHP framework is to using a technique suggested by Saaty (1977, 1980,
represent the problem in a hierarchical structure. 1982). (The calculation of a consistency index is

62 ©2002, AIMR®
 A Sensible Mutual Fund Selection Model

Figure 1. Decision Hierarchy for Mutual Fund Selection

Hierarchy 1: Selecting a Portfolio of Asset Classes

Determining a Suitable Set of

Asset Classes

Capital Appreciation Capital Appreciation Preservation of International

Current Income Tax Advantage
(5 years) (15 years) Principal Diversification

Municipal Growth and International Aggressive

Bonds
Bonds Income Equity Growth

Hierarchy 2: Selecting Mutual Funds from Each Asset Class

Growth and
Income

Manager’s
TR (15 years) Expense Ratio Sharpe Ratio
Tenure

Bartlett Basic ... IAI Growth & ... Washington

Value A Income Mutual Investors

Note: TR = total return.

presented in Appendix B.) Enforcing consistency is naire cannot identify such inconsistencies. No
an important contribution of the AHP. In tradi- financial advisor could recommend funds “suit-
tional financial planning, the investor indicates able” for this investor because of the inconsistent
preferences through a questionnaire administered preference ranking. Yet, the advisor could still be
by a financial advisor and the advisor interprets the held liable for recommendations in a lawsuit.
questionnaire subjectively; then, the advisor recom- If the advisor or investor observes inconsis-
mends funds that he or she believes to be suitable tency in the questionnaire answers, the investor can
to the investor.5 Unfortunately, investors seldom revise the answers until consistency is achieved.6
understand their own preferences fully and often
The advisor may not immediately know where
rank their preferences inconsistently. For example,
among the answers, however, the greatest inconsis-
suppose an investor ranks capital appreciation
tency lies. According to Saaty (1994a), “The AHP
(CA) as more important than current income (CI).
The investor also considers current income more can show one by one, in sequential order, which
important than preservation of principal (PP). The judgments are the most inconsistent, and also sug-
same investor then indicates that preservation of gests the value that best improves consistency” (p.
principal is more important than capital apprecia- 27). By providing the investor an opportunity to
tion. This investor has made three statements about reexamine preferences in a guided format, the AHP
preferences—CA > CI, CI > PP, PP > CA—that are enables the financial advisor and the investor to
inconsistent with each other. By itself, a question- better understand the investor’s true investment

May/June 2002 63
Financial Analysts Journal

Exhibit 1. Solving the Decision Hierarchy for Mutual Fund Selection

Hierarchy 1. Selecting a portfolio of asset classes
Investment objectives • Investor fills out a questionnaire based on the pairwise comparison scale developed by
Saaty (1980). (See sample for a hypothetical investor in Appendix A.)
• The consistency of investor’s responses is tested (Appendix B).
• If inconsistency is observed, investor can revise his/her answers until consistency is
achieved or method developed by Zeshui and Cuiping (1999) can be used to improve
consistency.
• Investor’s responses are transformed into relative-importance weights of investment
objectives, as in Table 1. (The calculation of relative-importance weights is discussed in
Appendix B.)

Major asset classes • Empirical data are used to obtain normalized performance scores that represent the
relative strengths of asset classes, as in Table 2. (The calculation of normalized
performance scores is presented in Appendix B.)

Ranking major asset classes • Relative-importance weights of investment objectives and relative strengths of asset
classes are used to calculate the ranking of asset classes, as in Table 3.

Hierarchy 2. Selecting mutual funds from each asset class

Fund selection criteria • Financial advisor determines the relative importance of fund selection criteria through
pairwise comparisons using the scale developed by Saaty (1980) and empirical research
on the relative importance of fund selection criteria.
• The consistency of financial advisor’s responses is tested (see Appendix B).
• If inconsistency is observed, financial advisor can revise his/her answers until
consistency is achieved or Zeshui–Cuiping method can be used to improve the
consistency.
• Financial advisor’s responses are transformed into relative-importance weights for fund
selection criteria, as in Table 4.

Fund performance under each criterion • Empirical data are used to obtain normalized performance scores that represent the
relative strengths of mutual funds, as in Table 5. (The calculation of normalized
performance scores is presented in Appendix B.)

Ranking mutual funds • Financial advisor uses relative-importance weights of fund selection criteria and relative
strengths of mutual funds to calculate the ranking of mutual funds under each asset class.
Table 6 presents the ranking of 10 randomly selected growth and income funds.

objectives. Alternatively, a method developed by After the consistency of the pairwise compari-
Zeshui and Cuiping (1999) can be used to improve son matrix has been verified, the next step is to
the consistency of an investor’s responses. Their estimate the relative-importance weight of each
method is based on an algorithm that takes an objective. Among a number of estimators, the
investor’s original responses and modifies them to eigenvalue method suggested by Saaty (1977, 1980,
achieve consistency without requiring the investor 1982) is widely used to obtain the relative weights.
to make any revisions. Zeshui and Cuiping pro- (A discussion of the eigenvalue method of comput-
vided examples showing that their algorithm is ing weight vectors from a pairwise comparison
convergent and practical. matrix is in Appendix B.) Panel B of Table 1 shows

Exhibit 2. Pairwise Comparison Scale

1 Equal importance Two attributes contribute equally to the objective.

3 Weak importance of one over another Experience and judgment slightly favor one attribute over another.

5 Essential or strong importance Experience and judgment strongly favor one attribute over another.

Very strong or demonstrated importance An attribute is favored very strongly over another; its dominance has
7 been demonstrated in practice.

Absolute importance The evidence favoring one attribute over another is of the highest
9 possible order of affirmation.

2, 4, 6, 8 Intermediate values between adjacent scale values When compromise is needed.

64 ©2002, AIMR®
 A Sensible Mutual Fund Selection Model

Table 1. Analysis of Pairwise Relative Importance of Objectives from Sample Questionnaire

Comparison Objectives
Capital Capital
Appreciation Appreciation Current Preservation International Tax
Preferred Objective (5 years) (15 years) Income of Principal Diversification Advantage
A. Investor’s responses to the sample questionnaire in a pairwise comparison format
Capital Appreciation (5 years) 1 3 1/4 1/3 3 1/3
Capital Appreciation
(15 years) 1/3 1 1/7 1/5 2 1/5
Current Income 4 7 1 4 8 3
Preservation of Principal 3 5 1/4 1 5 3
International Diversification 1/3 1/2 1/8 1/5 1 1/7
Tax Advantage 3 5 1/3 1/3 7 1

B. Relative importance of investment objectives

Weight 0.0866 0.0435 0.4339 0.2326 0.0324 0.1710

the relative-importance weights calculated by the was 1976–1980, and the second was 1977–1981. The
eigenvalue method for the investment objectives of value of CWI 5 in Panel A of Table 2 for an asset
the hypothetical investor. The most important class is the time-series average of the annual CWI 5
objective for this investor is current income, with a between 1976 and 1998.
weight of 0.4339. This investor also considers pres- The variable “cumulative wealth index for 15
ervation of principal and tax advantage more years” (CWI 15) represents the strength of an asset
important than capital appreciation and interna- class in serving the objective of capital appreciation
tional diversification. in 15 years. The calculation of the CWI 15 was the
The next step determines which asset classes same as that of the CWI 5 except the rolling period
are most suitable for the hypothetical investor. was 15 years.
The “current income” (CI) variable was com-
Ranking Asset Classes. The decision alter- puted as the percentage of income in a fund’s total
natives for the first hierarchy of the mutual fund annual return. We calculated the annual income
selection model are the asset classes. To make the component for each fund in an asset class for the
example more realistic, we have chosen five major 1976–98 period. And we used the overall average
asset classes—U.S. aggressive growth, U.S. growth as a proxy for measuring the strength of an asset
and income, U.S. bonds (corporate high-quality class in serving the current income objective.
and government bonds), municipal bonds, and We constructed the “index of negative devia-
international equity. We empirically estimated the tion” (IND) to represent the strength of an asset
suitability of the asset classes under each invest- class in serving the objective of preservation of
ment objective by using a sample from the Morn- principal. Because the focus is on downside risk in
ingstar database covering the 1976–98 period. By this objective, we considered only negative returns.
using empirical data, the mutual fund selection For each asset class, we averaged the annual nega-
model does not rely on the investor’s knowledge tive returns for the 1976–98 period. We considered
(or lack of knowledge) of finance in the decision the fund universe’s negative return to be the aver-
process. age negative return across all asset classes. The IND
To calculate each asset class’s relative strength for each asset class is the ratio of the asset class’s
in serving the investment objectives, we assigned average negative return to the fund universe nega-
proxies to the asset-class characteristics. We used a tive return. An IND value greater than 1 indicates
variable “cumulative wealth index for five years” the asset class has above-average downside risk
(CWI 5) as a proxy for the strength of an asset class compared with the fund universe.
in serving the objective of capital appreciation in We used the variable “percentage of fund
five years. The CWI 5 for each fund is the value of value in international assets” (PFVIA) to measure
a $1 investment by the end of five years. We com- the strength of an asset class in serving the interna-
puted the CWI 5 for each fund in the database for tional diversification objective. The overall average
each year and then used the cross-sectional average PFVIA for all funds in an asset class for 1976–1998
as the asset-class CWI 5. We repeated this process was used in the analysis.
for rolling five-year periods in the sample period The variable “tax advantage” (TA) was com-
(1976–1998); for example, the first five-year period puted as the ratio of three-year average tax-

May/June 2002 65
Financial Analysts Journal

Table 2. Asset Class Relative Strengths in Serving Investment Objectives

Asset class CWI 5 CWI 15 CI IND PFVIA TA
A. Historical (1976–98) average of investment objective attributes in class
Aggressive growth $1.8298 $10.4296 0.97% 1.7525 2.90% 0.8555
Bonds a 1.0183 1.0906 8.52 0.3478 0.00 0.5883
International equity 1.5973 4.6165 1.79 1.1929 89.90 0.8145
Growth and income 1.5134 5.4107 3.39 1.0044 3.70 0.8709
Municipal bonds b 1.0296 1.0300 6.81 0.4879 0.00 0.9804

B. Relative-strength weight of class under objective

Aggressive growth 0.3798 0.4538 0.0442 0.0362 0.0926 0.2194
a
Bonds 0.0422 0.0530 0.3979 0.3255 0.0736 0.0650
International equity 0.2830 0.2043 0.0827 0.1514 0.6624 0.1957
Growth and income 0.2481 0.2384 0.1574 0.1903 0.0978 0.2283
Municipal bonds b 0.0469 0.0504 0.3178 0.2966 0.0736 0.2916
a Bonds
are corporate high-quality bond funds and government bond funds.
b Municipal bonds are national municipal bond funds.

adjusted total return to three-year average total Table 3. Asset-Allocation Recommendation for
return for each asset as of 1998. The TA variable the Hypothetical Investor
measures the strength of an asset class in serving Asset Class Weight
the objective of tax advantage.
Bonds 0.2678
Panel A of Table 2 shows the values of the
Municipal bonds 0.2654
variables used as proxies for the suitability of asset
Growth and income 0.1866
classes for each investment objective. International equity 0.1594
Next, we normalized these values to obtain the Aggressive growth 0.1208
relative-strength weights of the asset classes under
each investment objective. (A discussion of the nor- tage. Therefore, the asset-allocation recommenda-
malization method is in Appendix B.) The relative- tion in Table 3 assigns bonds the highest weight,
strength weights of the asset classes in serving each followed by municipal bonds, with aggressive
objective are presented in Panel B of Table 2. For growth receiving the lowest weight. The asset-
example, under the objective of tax advantage (TA), allocation recommendation in Table 3 is not only
municipal bonds have the largest weight, indicat- based on empirical data but also takes into account
ing that municipal bonds are the most suitable the relative importance the investor gave to each
investment for this objective. investment objective.
Once the relative importance of investment
objectives and strength of each asset class’s contri- Selecting Individual Mutual Funds. Once
bution to each objective have been determined, the relative importance of the asset classes has been
they are combined to obtain the asset-allocation determined, the next step involves the second hier-
weights suitable for the investor. For the hypothet- archy—evaluating individual mutual funds within
ical investor, we used the strength of the asset each class. Hierarchy 2 in Figure 1 shows the rank-
classes under each investment objective (given in ing of individual mutual funds in the growth and
the 5 × 6 matrix in Panel B of Table 2) and put the income asset class. 7 The mission, the top level of the
relative-importance weights of the investment second hierarchy, is to obtain a ranking of growth
objectives to the hypothetical investor (as given in and income funds. The selection criteria, the second
Table 1) into a 6 × 1 vector. The asset-allocation level of the hierarchy, are based on the fund struc-
weights (a 5 × 1 vector) are the product of the tural and operational characteristics.8 The available
strength matrix (5 × 6) and the relative-importance alternatives, the third level, are all the mutual funds
vector (6 × 1). Table 3 presents the resulting asset- from the database in the growth and income asset
allocation weights for the hypothetical investor. class.
Recall from Table 1 that the most important ■ Selection criteria. The AHP mutual fund
objectives for this hypothetical investor are current selection model uses four criteria to rank mutual
income, preservation of principal, and tax advan- funds: the total return over 15 years annualized (TR

66 ©2002, AIMR®
 A Sensible Mutual Fund Selection Model

15), the expense ratio, the fund manager’s tenure, selection criteria for growth and income funds is
and the Sharpe ratio (excess return divided by the presented in Panel B of Table 4.
standard deviation of return). These criteria were The weights for the TR 15 and the Sharpe ratio
chosen on the basis of research findings. As for imply that, in seeking to match funds with investor
performance, on the one hand, previous studies preferences, the financial advisor places equal
(Grinblatt and Titman; Hendricks, Patel, and Zeck- emphasis on a fund’s long-term return perfor-
hauser 1993) have shown that funds with the low- mance and its reward-to-risk performance. The
est returns tend to continue to underperform. On expense ratio comes next in relative importance.
the other hand, whether superior performance is ■ Fund performance under each criterion. The
persistent remains an unresolved issue (Malkiel; next step in the decision hierarchy is to use the
Elton et al.; Carhart). Including past performance selection criteria to evaluate individual funds from
(the TR 15 and Sharpe ratio) as selection criteria was the growth and income asset class. For illustrative
intended to weed out funds with the lowest purposes, we randomly chose 10 growth and
returns. A fund’s expense ratio has been shown to income funds as available funds. Panel A of Table
be negatively related to performance in bond funds 5 presents the TR 15, the most recent expense ratio,
(Blake et al.; Detzler) and in stock funds (Carhart). manager’s tenure, and the Sharpe ratio for these 10
Finally, many financial advisors consider the fund funds. Panel B of Table 5 transforms the data in
manager’s tenure an important attribute for esti- Panel A into relative-strength weights by normal-
mating future performance. izing the values (see Appendix B). Panel A shows
that Dodge & Cox has the lowest expense ratio and
The relative importance of these four criteria
Philadelphia has the highest expense ratio. Because
may differ among asset classes. For example,
a low expense ratio is desirable, Dodge & Cox has
expense ratio is a more important factor for bond
a relative strength of 0.1572 whereas Philadelphia
funds than for equity funds. Given the complexity
has a relative strength of only 0.0175 under this
of the issues, the task of assigning relative impor-
criterion. For past performance, high return is pre-
tance should be performed by a financial advisor.
ferred, so under the TR 15 criterion, Washington
The advisor may carry out a pairwise comparison
Mutual Investors, which has the highest TR 15, has
of fund selection criteria via a questionnaire similar the highest relative strength.
to the sample provided in Appendix A, or the ■ Ranking mutual funds. The last level in the
advisor may rely on research findings in forming fund selection hierarchy is the ranking of the
his/her judgments about the relative importance of mutual funds. For the example growth and income
fund selection criteria. Panel A of Table 4 shows asset class, to facilitate the computation, we put the
the pairwise comparisons of the four criteria for relative-strength weights from Panel B of Table 5
growth and income funds assigned by a hypothet- into a 10 × 4 matrix. The ranking is the product of
ical financial advisor. Once the consistency of the the matrix of relative-strength weights and the rel-
pairwise comparison matrix has been verified, the ative importance of the criteria (given in Panel B of
vector of relative-importance weights can be calcu- Table 4). Table 6 presents the rankings of the 10
lated using the eigenvalue methodology. The randomly chosen growth and income funds. Given
resulting relative-importance weight vector of the the importance assigned to past returns (TR 15 and
the Sharpe ratio), it is not surprising that funds with
high returns, such as Washington Mutual Investors
Table 4. Analysis of Relative Importance of Fund and MFS Massachusetts Inv. A, received high rank-
Selection Criteria for Growth and ings. The rankings take into account both the his-
Income Funds torical strength of the funds as related to each
criterion and the relative importance of each crite-
Comparison Criterion
rion. Thus, the rankings generated by the AHP
Preferred Expense Manager’s Sharpe
model are a more comprehensive evaluation metric
Criterion TR 15 Ratio Tenure Ratio
than risk-adjusted performance measures.
A. Pairwise comparison of criteria
After the AHP model has determined the asset-
TR 15 1 3 5 1
allocation recommendation and the rankings for
Expense Ratio 1/3 1 3 1/3
funds within each asset class, it is up to the financial
Manager’s
Tenure 1/5 1/3 1 1/5
advisor and the investor to decide how many funds
Sharpe Ratio 1 3 5 1
from each asset class to include in the final portfo-
lio. For example, O’Neal (1997) found that five
B. Relative importance of criteria funds are needed to create a well-diversified port-
Weight 0.39 0.15 0.07 0.39 folio of growth funds.

May/June 2002 67
Financial Analysts Journal

Table 5. Analysis of the Relative Strengths of 10 Randomly Selected Growth

and Income Funds under Fund Selection Criteria
Expense Manager’s Sharpe
Fund TR 15 Ratio Tenure Ratio
A. Most recently available data on fund as of 1998
Bartlett Basic Value A 12.90% 1.13% 16 years 0.73
Bridges Investment 12.91 0.81 36 1.06
Dodge & Cox Stock 16.38 0.57 21 0.86
Hancock Growth & Income A 14.81 1.12 3 1.09
IAI Growth and Income 13.34 1.25 3 0.88
MFS Massachusetts Inv. A 16.81 0.74 6 1.31
Philadelphia 11.76 1.53 12 1.01
Sit Large-Cap Growth 16.07 1.00 17 1.13
UMB Scout Stock 11.85 0.86 17 0.69
Washington Mutual Investors 17.13 0.62 10 1.30

B. Relative-strength weight of fund

Bartlett Basic Value A 0.0548 0.0757 0.1125 0.0299
Bridges Investment 0.0550 0.1223 0.2438 0.1137
Dodge & Cox Stock 0.1600 0.1572 0.1453 0.0629
Hancock Growth & Income A 0.1125 0.0771 0.0271 0.1213
IAI Growth and Income 0.0681 0.0582 0.0271 0.0680
MFS Massachusetts Inv. A 0.1730 0.1325 0.0468 0.1773
Philadelphia 0.0203 0.0175 0.0862 0.1010
Sit Large-Cap Growth 0.1506 0.0946 0.1190 0.1315
UMB Scout Stock 0.0230 0.1150 0.1190 0.0197
Washington Mutual Investors 0.1827 0.1499 0.0731 0.1747

Table 6. Ranking of Growth and Income Funds is not intended to create a portfolio to provide
Relative-
superior future performance alone but to ensure
Importance AHP Model that the asset allocation is consistent with the objec-
Mutual Fund Weight Ranking tives and preferences of the investor.
Washington Mutual Investors 0.167 1 A recommendation of specific mutual funds,
MFS Massachusetts Inv. A 0.160 2 the final objective of the AHP model, is the most
Sit Large-Cap Growth 0.133 3 important function of a financial advisor. Most
Dodge & Cox Stock 0.121 4 financial advisors believe that, for this function,
Hancock Growth & Income A 0.105 5 mutual funds should be evaluated on other
Bridges Investment 0.101 6 attributes in addition to risk and return. But the
IAI Growth and Income 0.064 7 process they follow of incorporating other
Philadelphia 0.056 8 attributes, especially qualitative attributes, is often
Bartlett Basic Value A 0.052 9 ad hoc, and the overall evaluation process appears
UMB Scout Stock 0.042 10 to be a “black box” to the investor. An advantage
of the AHP approach is that it helps advisors
Summary and Implications explain the fund selection process to investors. It
We have recommended the analytic hierarchy pro- requires the advisor to list the selection criteria and
cess as a method to solve the complex decision the advisor’s beliefs about the relative importance
problem of selecting mutual funds. The asset allo- of each criterion. The final rankings generated by
cations recommended by the AHP model are cus- the AHP model can be shown to incorporate the
tomized to fit the investment objectives and funds’ strength according to each criterion and the
constraints of a particular investor. The AHP model relative importance of each criterion.

68 ©2002, AIMR®
 A Sensible Mutual Fund Selection Model

Appendix A. Sample Questionnaire Specify the relative importance using the fol-
lowing scale.
Note: Responses from the hypothetical investor are
1 2 3 4 5 6 7 8 9
in bold italic.
8. Which is more important: capital appreciation
Use a scale of 1–9, where 1 indicates both objectives over the next 15 years versus international
are equally important and 9 indicates the one objec- diversification?
tive is absolutely more important, to compare the Capital appreciation over the next 15 years.
relative importance of each investment objective.
Specify the relative importance using the fol-
1. Which is more important: capital appreciation
lowing scale.
over the next 5 years versus capital appreciation
1 2 3 4 5 6 7 8 9
over the next 15 years?
9. Which is more important: capital appreciation
Capital appreciation over the next 5 years. over the next 15 years versus tax advantage?
Specify the relative importance using the fol- Tax advantage.
lowing scale.
Specify the relative importance using the fol-
1 2 3 4 5 6 7 8 9
lowing scale.
2. Which is more important: capital appreciation 1 2 3 4 5 6 7 8 9
over the next 5 years versus current income? 10. Which is more important: current income ver-
Current income. sus preservation of principal?
Specify the relative importance using the fol- Current income.
lowing scale. Specify the relative importance using the fol-
1 2 3 4 5 6 7 8 9 lowing scale.
3. Which is more important: capital appreciation 1 2 3 4 5 6 7 8 9
over the next 5 years versus preservation of 11. Which is more important: current income ver-
principal? sus international diversification?
Preservation of principal. Current income.
Specify the relative importance using the fol- Specify the relative importance using the fol-
lowing scale. lowing scale.
1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9
4. Which is more important: capital appreciation 12. Which is more important: current income ver-
over the next 5 years versus international diver- sus tax advantage?
sification? Current income.
Capital appreciation over the next 5 years. Specify the relative importance using the fol-
lowing scale.
Specify the relative importance using the fol-
1 2 3 4 5 6 7 8 9
lowing scale.
1 2 3 4 5 6 7 8 9 13. Which is more important: preservation of prin-
cipal versus international diversification?
5. Which is more important: capital appreciation
over the next 5 years versus tax advantage? Preservation of principal.
Specify the relative importance using the fol-
Tax advantage.
lowing scale.
Specify the relative importance using the fol- 1 2 3 4 5 6 7 8 9
lowing scale.
14. Which is more important: preservation of prin-
1 2 3 4 5 6 7 8 9
cipal versus tax advantage?
6. Which is more important: capital appreciation Preservation of principal.
over the next 15 years versus current income?
Specify the relative importance using the fol-
Current income. lowing scale.
Specify the relative importance using the fol- 1 2 3 4 5 6 7 8 9
lowing scale. 15. Which is more important: international diversi-
1 2 3 4 5 6 7 8 9 fication versus tax advantage?
7. Which is more important: capital appreciation Tax advantage.
over the next 15 years versus preservation of Specify the relative importance using the fol-
principal? lowing scale.
Preservation of principal. 1 2 3 4 5 6 7 8 9

May/June 2002 69
Financial Analysts Journal

Appendix B. Model Calculations CI

CR = ----------- 100, (B5)
ACI
Once the inputs to the pairwise comparison matrix
have been obtained, the relative-importance where ACI is the average consistency index of ran-
weights of the compared alternatives and the con- domly generated relative-weight matrixes. Saaty
sistency of the pairwise comparisons are calculated (1980) suggested that a CR value of 10 percent or
through use of a technique suggested by Saaty less is considered acceptable. Otherwise, it is rec-
(1977, 1980, 1982). These calculations are based on ommended that the decision maker reconstruct P̂
the following discussion. to resolve inconsistencies in the pairwise compari-
sons.
Weight Vector. If the decision maker knew
the actual relative weights, w, of n elements, the Normalizing Performance Scores. We nor-
matrix of pairwise comparisons, P, would be malized the values of variables that represent the
relative strengths of asset classes in serving each
w1 ⁄ w1 w1 ⁄ w2 … w1 ⁄ wn investment objective and the relative strengths of
w2 ⁄ w1 w2 ⁄ w2 … w2 ⁄ wn mutual funds in serving each fund selection crite-
rion. First, we transformed the values so that they
P = · · · · (B1) map onto the interval ranging from 1 to 9, where
· · · ·
the minimum value and the maximum value are
· · · · transformed to 1 and 9, respectively. The variables
wn ⁄ w 1 w n ⁄ w2 … w n ⁄ wn for which the decision maker requires smaller val-
ues were transformed so that the maximum value
and the relative weights could be obtained from
corresponds to 1 and the minimum value corre-
each one of n rows of matrix P. Hence, matrix P has
sponds to 9. Then, we normalized the transformed
rank 1 and
values to unity to obtain the relative-strength
P(W) = n(W), (B2) weights.9
where W = (w1, w2, . . ., wn)T
is the vector of actual The following algorithm describes this process
relative weights and n is the number of elements; n with an example. The values for CWI 5, which
is the eigenvalue of matrix P, and W is the right come from Panel A of Table 2, are
eigenvector of matrix P. a = (1.8298, 1.0183, 1.5973, 1.5134, 1.0296). (B6)
The AHP postulates that the decision maker We transformed each value by using
does not know W and, therefore, is not able to
produce the pairwise comparison elements of a i – a min
a′i = 1 + --------------------- , (B7)
matrix P accurately. Consequently, the matrix gen- L
erated by the decision maker contains inconsisten-
where L represents the step length as calculated in
cies. An estimate of W, which is denoted by Ŵ , can
Equation B8:
be obtained, however, by using the following equa-
tion: a max – a min
L = ---------------------------- . (B8)
9–1
P̂ ( Ŵ ) = λ max ( Ŵ ), (B3)
Equation B9 shows the transformed values
where P̂ is the observed matrix of pairwise com-
parisons, λmax is the largest eigenvalue of P̂ , and a′ = ( 9, 1 , 6.7076 , 5.8807, 1.1115 ). (B9)
Ŵ is its right eigenvector. The variable λmax is The transformed values were normalized by using
considered an estimate of n in Equation B2.
a′i
α i = ------------
∑
- . (B10)
Consistency Testing. The consistency of the a′i
i
pairwise comparisons is tested by using the consis-
tency index and the consistency ratio. The consis- The normalized values are
tency index (CI) is based on the premise that the α = (0.3798, 0.0422, 0.2830, 0.2481, 0.0469). (B11)
closer the value of λmax is to n, the more consistent
If the decision maker requires a minimum
are the values of P̂ :
value for a particular variable, such as IND (the
( λ max – n ) index of negative deviation), Equation B12 can be
CI = -------------------------- . (B4) used to transform the values:
(n – 1 )

With the use of the CI, the consistency ratio a i – a min

a′i = 9 – --------------------- . (B12)
(CR) can be calculated as L

70 ©2002, AIMR®
 A Sensible Mutual Fund Selection Model

Notes
1. Eliminating internal inconsistencies among the investor’s 6. Saaty (1994b) argued that revising judgments is an accept-
investment preferences is especially valuable to financial able and, in fact, essential process in the AHP.
advisors because it allows them to demonstrate that they 7. For brevity, we limited our performance analysis to the
have exercised due diligence in assessing the investor’s growth and income asset class. A similar analysis would
goals and risk tolerance in an objective manner. apply to each of the other classes.
2. A search using “mutual funds” as the key words located 8. Note that in the sample model (Figure 1), some of the
105 books at Amazon.com and more than 16,000 articles selection criteria in the first hierarchy are not included in
since 1988 at InfoTrak. the second hierarchy. The model assumes that all the funds
3. The items are risk level, total return, investment goals, types in an asset class have similar attributes for the omitted
of company in which the fund invests, annual fees, sales criteria (e.g., preservation of capital). If funds in an asset
charge, fund manager’s investment style, total assets, port- class are not homogeneous with respect to an omitted cri-
folio turnover rate, fund manager’s professional back- terion, then that category should be included in the second
ground, and an 800 telephone number for customers. hierarchy.
4. For example, participants in a company pension plan may 9. Similar normalization methods were used in studies that
have unique constraints specific to the sponsoring com- incorporated quantitative data in the AHP (Yu, Jin, Zhang,
pany. Ling, and Barnes 2000; Weck, Klocke, Schell, and Ruenau-
5. Investors who do not use a professional advisor often fail ver 1997).
to establish investment objectives and select funds on an ad
hoc basis.

References
Arbel, A., and A. Seidmann. 1984. “Selecting a Microcomputer Johnson, Ronald A., V. Srinivasan, and P.J. Bolster. 1990.
for Process Control and Data Acquisition.” IIE Transactions, vol. “Sovereign Debt Ratings: A Judgmental Model Based on the
16, no. 1 (March):73–80. Analytic Hierarchy Process.” Journal of International Business
Blake, C., E. Elton, and M. Gruber. 1993. “The Performance of Studies, vol. 21, no. 1 (First Quarter):95–117.
Bond Mutual Funds.” Journal of Business, vol. 66, no. 3 (July):371– Khaksari, Shahriar, R. Kamath, and R. Grieves. 1989. “A New
403. Approach to Determining Optimum Portfolio Mix.” Journal of
Bogle, John C. 1994. Bogle on Mutual Funds. Burr Ridge, IL: Irwin. Portfolio Management, vol. 15, no. 3 (Spring):43–49.

Bolster, P.J., and S. Warrick. 1999. “Suitability and Optimality in Malkiel, B. 1995. “Returns from Investing in Equity Mutual
the Asset Allocation Process: Conflict and Resolution Using the Funds: 1971 to 1991.” Journal of Finance, vol. 50, no. 2 (June):549–
Analytic Hierarchy Process and Mean Variance Optimization.” 572.
Working paper, Northeastern University. Merrill Lynch Financial Literacy Index Analysis: A Nationwide
Survey of Adults. 1994. Luntz Research.
Bolster, P.J., V. Janjigian, and E.A. Trahan. 1995. “Determining
Investor Suitability Using the Analytic Hierarchy Process.” Miller, G.A. 1956. “The Magical Number Seven Plus or Minus
Financial Analysts Journal, vol. 51, no. 4 (July/August):63–75. Two: Some Limits on Our Capacity for Processing Information.”
Psychological Review, vol. 63, no. 2 (March):81–97.
Brennan, M. 1995. “The Individual Investor.” Journal of Financial
Research, vol. 18, no. 1 (Spring):59–74. Mutual Fund Fact Book. 1997. Washington DC: Investment
Company Institute.
Ca rh art, M. 19 97. “On Persistence i n M utual Fund
Performance.” Journal of Finance, vol. 52, no. 1 (March):57–82. ———. 2000. Washington DC: Investment Company Institute.
Detzler, M. 1999. “The Performance of Global Bond Mutual Olsen, R.A. 1998. “Behavioral Finance and Its Implications for
Funds.” Journal of Banking and Finance, vol. 23, no. 8 (July):1195– Stock-Price Volatility.” Financial Analysts Journal, vol. 54, no. 2
1217. (March/April):10–18.
Elton, E., M. Gruber, and C. Blake. 1996. “The Persistence of O’Neal, E. 1997. “How Many Mutual Funds Constitute a
Risk-Adjusted Mutual Fund Performance.” Journal of Business, Diversified Mutual Fund Portfolio?” Financial Analysts Journal,
vol. 69, no. 2 (January):133–157. vol. 53, no. 2 (March/April):37–46.
Eun, Cheol, R. Kolodny, and B. Resnick. 1991. “U.S.-Based Puelz, Robert. 1991. “A Process for Selecting a Life Insurance
International Mutual Funds: A Performance Evaluation.” Contract.” Journal of Risk and Insurance, vol. 58, no. 1
Journal of Portfolio Management, vol. 17, no. 3 (Spring):88–94. (March):138–146.
Grinblatt, M., and S. Titman. 1992. “The Persistence of Mutual Saaty, T.L. 1977. “A Scaling Method for Priorities in Hierarchical
Fund Performance.” Journal of Finance, vol. 47, no. 5 Structures.” Journal of Mathematical Psychology, vol. 15, no. 3
(December):1977–84. (June):234–281.

Hendricks, D., J. Patel, and R. Zeckhauser. 1993. “Hot Hands in ———. 1980. The Analytic Hierarchy Process. New York: McGraw-
Mutual Funds: Short-Run Persistence of Relative Performance, Hill.
1974–1988.” Journal of Finance, vol. 48, no. 1 (March):93–130. ———. 1982. Decision Making for Leaders: The Analytic Hierarchy
Johnson, C.R., and J.M. Hihn. 1980. “Prioritization of Research Process for Decisions in a Complex World. Atlanta, GA: Lifetime
Projects.” In Energy Storage: The Transactions of the First Learning Publications.
International Assembly on Energy Storage. Edited by J. Silverman. ———. 1994a. “How to Make a Decision: The Analytic Hierarchy
New York: Pergamon:562–566. Process.” Interfaces, vol. 24, no. 6 (November/December):19–43.

May/June 2002 71
Financial Analysts Journal

———. 1994b. “The Analytic Hierarchy Process: Some Yu, Yue, Kai Jin, Hong C. Zhang, Frederick F. Ling, and David
Observations on the Paper by Apostolou and Hassell.” Journal Barnes. 2000. “A Decision-Making Model for Materials
of Accounting Literature, vol. 13:212–218. Management of End-of-Life Electronic Products.” Journal of
Sinuany-Stern, Z. 1984. “A Network Optimization Model for Manufacturing Systems, vol. 19, no. 2 (July):94–107.
Budget Allocation in a Multicampus University.” Journal of the Zahedi, Fatemeh. 1986. “The Analytic Hierarchy Process: A
Operational Research Society, vol. 35, no. 8 (August):749–757. Survey of the Method and Its Applications.” Interfaces, vol. 16,
Weck, M, F. Klocke, H. Schell, and E. Ruenauver. 1997. no. 4 (July/August):96–108.
“Evaluating Alternative Production Cycles Using the Extended Zeshui, Xu, and Wei Cuiping. 1999. “A Consistency Improving
Fuzzy AHP Method.” European Journal of Operational Research, Method in the Analytic Hierarchy Process.” European Journal of
vol. 100, no. 2 (July):351–366. Operational Research, vol. 116, no. 2 (July):443–449.

72 ©2002, AIMR®