Explaining Mutual Fund Performance

Kenneth Khang†
College of Business
Idaho State University
Pocatello, ID 83209
Ph: (208) 282-6398
e-mail: khankenn@isu.edu

Thomas W. Miller, Jr.

College of Business
Mississippi State University
P.O. Box 9580
Mississippi State, MS 39762
Ph: (662) 325-1997
e-mail: tmiller@business.msstate.edu

May 2019
JEL Classifications: G12; G14; G23
Keywords: Portfolio Performance Measurement; Portfolio Performance Analysis; Unconditional
Performance; Morningstar Mutual Funds; Asset Allocation

Part of this work was done when Miller was at the John Cook School of Business, Saint Louis
University. We thank Leigh Riddick for providing weight allocation data and acknowledge Mary
Bange for her contributions on related collaborations. We are grateful for helpful comments and
suggestions from Fan Chen, Rohan Christie-David, Ken Cyree, David Dubofsky, Heber
Farnsworth, Mike Faulkender, Wayne Ferson, Kathleen Fuller, Mike Hemler, Mike Highfield,
Carolyn Moore, Gideon Ozik, Jayendu Patel, Oleg Rytchkov, Steven Thorley, Bonnie Van Ness,
Robert Van Ness, and seminar participants in the Department of Agricultural Economics,
Mississippi State University and at the University of Mississippi, the University of Louisville, and
the 2009 FMA International Conference. Portions of this paper in early drafts were presented in
seminars at Mississippi State University and Saint Louis University, where we received many
helpful suggestions. Of course, we are responsible for any remaining errors.
†Corresponding Author.
 Explaining Mutual Fund Performance

Abstract
We use an empirically tractable method to disentangle portfolio returns into components and use
those components to explain the performance of a sample of 732 asset allocation mutual funds
during the period December 1983 through June 2013. We separate a portfolio’s Active Returns, or
returns in excess of a benchmark, into three measurable components that we call Forecasting
Success, Active Management, and Return Dispersion. Forecasting Success reflects a manager’s
success at forecasting asset returns over a future period. Active Management indicates how
aggressively a manager alters portfolio weights. Return Dispersion reflects the degree of asset
return dispersion. We find that the average mutual fund does not beat a passive lagged-weight
benchmark, but do find cross-sectional dispersion in fund performance across the sample that is
explained by the components. We then examine the persistence of Active Returns and of each
component. We find no persistence in Active Returns, indicating that the distribution of fund
Active Returns is generated by chance, which is consistent with Fama and French (2010). We find
strong persistence in the Active Management component only. This persistence in Active
Management indicates that funds with high Active Management in the past will continue to exhibit
high Active Management in the future, which coincides with more extreme (both good and bad)
performance. We also find evidence that past negative Forecasting Success leads to greater
subsequent Active Management, which is consistent with funds responding to negative past Active
Returns by making larger bets in the future, perhaps in an attempt to try to overcome past negative
Active Returns.
 Explaining Mutual Fund Performance

I. Introduction

Measuring portfolio performance has long been of interest to Finance academics and

practitioners alike. Most of the literature focuses on examining portfolio returns exclusively (eg.

measuring an alpha or Sharpe ratio). A more recent strand of the literature focuses on directly

examining the actions of portfolio managers and then measuring whether these actions affect

portfolio returns. Wermers (2003) was the first to consider such active portfolio management as a

measureable quantity. Lo (2008) separates portfolio returns into active and passive returns.

Cremers and Petajisto (2009) examine mutual fund managers and create a statistic to describe their

activity level.

We add to this literature by implementing a simple separation of returns into components

that directly disentangles the effects on portfolio returns of observable actions made by portfolio

managers. In this paper, we use a weights-based approach to measure active portfolio returns (AR),

or returns in excess of a benchmark 1. This leads us to a method to characterize active portfolio

returns (AR) as a combination of three measurable and intuitive components that we call

Forecasting Success (FS), Active Management (AM), and Return Dispersion (RD). More

specifically, over a specific period AR = N * FS * AM * RD, where N is the number of assets in

the portfolio.

Forecasting Success (FS) reflects how well a manager forecasts relative asset returns over

a future period. Active Management (AM) reflects how active a manager is in altering the

1
We follow Lo (2008) in separating a portfolio’s return into an active component, or which is the return in excess of
a benchmark, and a passive component, which is the benchmark return.

1
portfolio’s composition. Return Dispersion (RD) reflects the amount of relative asset return

dispersion. We develop empirical measures of these components and then apply them to analyze

a sample of asset allocation mutual funds. We find significant cross-sectional Active Return (AR)

dispersion across funds, but no evidence that the average asset allocation mutual fund outperforms

a lagged weight benchmark.

While much of the academic literature supports the notion that mutual funds on average do

not generate positive active returns (see Fama and French (2010)), there is some evidence

indicating there are subsets of mutual funds that may generate positive active returns. Wermers

(2000) finds that returns of high turnover mutual funds exceed the returns of the Vanguard 500

Index Fund after expenses and transaction costs. Kacperczyk, Sialm, and Zheng (2005) find

evidence that equity mutual fund managers who exhibit industry concentration in their portfolios

outperform common equity benchmarks. Cremers and Petajisto (2009) show that equity mutual

funds with more active management perform better than those with less active management. These

results imply that managers with Forecasting Success (FS) in selecting equities also exhibit greater

Active Management (AM) by holding portfolios that stray further from common benchmarks.

Unlike the papers mentioned above that examine equity mutual funds, we implement our

portfolio measures in the realm of active asset allocation across broad asset classes. This allows us

to avoid the pesky distinction between market timing and individual security selection (see

Cremers and Petajisto (2009) among others) and provides a look at a segment of the mutual fund

universe that is often overlooked. Because we apply our method to active asset allocation mutual

funds, our paper has important implications for investors who rely on asset allocation advice. Our

paper also shows how examining Forecasting Success (FS), Active Management (AM), and Return

Dispersion (RD) can provide additional insight into mutual fund performance.

2
 Finally, our method requires only a history of beginning-of-period portfolio weights and

the subsequent periodic returns on the assets used. We consider active portfolio managers whose

asset allocation consists mainly of three assets: equities, bonds, and cash. Our method, however,

can be modified to examine other portfolio types, including equity funds, bond funds, hedge funds,

and other alternative investments funds. However, we note that a focus on asset allocation funds

is not trivial. As of the end of 2017, about 8 percent of the $18.7 trillion in net assets in mutual

funds world-wide were in the Hybrid, or the Asset Allocation/Balanced, category. 2

Our paper proceeds as follows. Section II contains a brief discussion of related research. In

Section III, we motivate and define the components of Forecasting Success (FS), Active

Management (AM), and Return Dispersion (RD). In Section IV, we present the empirical measures

of Active Returns (AR), Forecasting Success (FS), Active Management (AM), and Return

Dispersion (RD). In Section V, we apply our empirical measures to a sample of active asset

allocation mutual funds. Section VI contains our summary and conclusions.

II. Related Research

A. Weights-Based Portfolio Performance Measurement

Weights-based portfolio performance measures are based on the simple idea that skilled

portfolio managers will shift their portfolio weights in the same direction as subsequent returns. 3

To test this relationship, Copeland and Mayers (1982) and Grinblatt and Titman (1993) develop

unconditional weight-based portfolio performance measures. Conceptually, these measures equate

a portfolio manager’s performance with the unconditional covariance between her portfolio

weights and subsequent returns.

2
ICI 2018 Investment Company Fact Book from www.ici.org
3
Wermers (2006) provides a detailed survey of weights-based portfolio performance measurement.

3
 Ferson and Khang (2002) create a weights-based portfolio performance measure that

disentangles public information from a manager’s private information. They show that it is

possible for a manager to exhibit positive unconditional performance, while having no conditional

performance. This can occur if a portfolio manager uses publicly available information when

purchasing and selling securities. The performance measure used in this paper generates portfolio

manager rankings that reflect their unconditional performance. While it is possible to create

conditional versions of our measures, we do not do this to avoid having to specify the predictor

variables in the public information set for equity, bond, and cash markets.

B. Previous Measures of Active Management

Despite its importance to academics and practitioners, research on measuring and/or

characterizing active management is relatively scarce. Wermers (2003) was the first to consider

active management as a measureable quantity. He uses the S&P 500 tracking error as a measure

of active management and finds that U.S. domestic equity mutual funds with greater tracking error

perform better. Wermers (2003) concludes the managers of those funds appear to have better stock-

picking skills.

Cremers and Petajisto (2009) introduce a concept they call Active Share, AS, as a measure

of active management, which has recently become popular in the Investments industry. 4 They

examine equity mutual funds and find a positive relationship between AS and alpha from the four-

factor model of Carhart (1997). They also find significant variation in the degree of manager

activity among funds. We emphasize the AS measure of Cremers and Petajisto (2009) and our

4
See CFA Program Curriculum 2017 Level 2 Volume 6 on page 453 by Clark, de Silva, and Thorley (2017) and an
article by John Rekenthaler from http://www.morningstar.com/advisor/t/92563289/active-share-what-you-need-to-
know.htm

4
measure of Active Management, AM, indicate only the activity level of a manager, and not whether

the activity resulted in good or bad outcomes. Specifically, both the AS and AM simply measure

the size of weight deviations from the benchmark over a measurement period. To determine

whether the activity produced was beneficial or detrimental, they use alpha, while we use average

Active Returns (AR). We also show that Forecasting Success (FS) determines whether an increase

in manager activity will be beneficial or detrimental.

III. Measuring Performance and its Components

The focus of our paper is to simultaneously measure the performance of active managers

and break down that performance to isolate the effects of a manager’s observable actions, which

we show are contained in Forecasting Success (FS) and Active Management (AM). This allows

one to rank portfolio managers based on their success in adjusting their portfolio weights toward

those assets that will have higher returns over a future period and based on their activity level.

A. Portfolio Performance

Consider the following definition of portfolio performance, or Active Returns (AR), over

period t:

𝐴𝐴𝐴𝐴𝑝𝑝,𝑡𝑡 ≡ 𝑟𝑟̃𝑝𝑝,𝑡𝑡 − 𝑟𝑟̃𝑏𝑏,𝑡𝑡 (1).

In Equation (1), 𝑟𝑟̃𝑝𝑝,𝑡𝑡 is the period t return of a portfolio and 𝑟𝑟̃𝑏𝑏,𝑡𝑡 is the period t return of a benchmark

portfolio. Thus, Active Return (AR) is simply how much a portfolio beats its benchmark over some

time period.

Later, when we implement Equation (1) empirically in this paper, the benchmark portfolio

will be comprised of the lagged portfolio weights. This allows us to avoid having to specify

5
individual mutual fund benchmarks and mismatching a portfolio with an inappropriate benchmark.

Also, it allows us to avoid having to adjust for differences in expected asset returns, since the

benchmark uses the same assets in similar proportions. It is important to note, however, that the

benchmark portfolio in Equation (1) can be whatever an investor or client desires. For example, a

client may task a money manager with outperforming a weight allocation of 60-30-10 to equities,

bonds, and cash. Note a client would likely need to impose some risk controls on such a

benchmark, since a 100% allocation to equities would likely outperform based on historical

averages. Using the lagged portfolio weights as our benchmark greatly mitigates this issue, as we

will show.

Inserting the portfolio weights and taking expectations, we can rewrite Equation (1) as

𝑏𝑏
E�𝑟𝑟̃𝑝𝑝,𝑡𝑡 − 𝑟𝑟̃𝑏𝑏,𝑡𝑡 � = 𝐸𝐸�∑𝑁𝑁 �𝑗𝑗,𝑡𝑡 𝑟𝑟̃𝑗𝑗,𝑡𝑡 − ∑𝑁𝑁
𝑗𝑗=1 𝑤𝑤 𝑗𝑗=1 𝑤𝑤𝑗𝑗,𝑡𝑡 𝑟𝑟̃𝑗𝑗,𝑡𝑡 � (2).

In Equation (2), 𝑤𝑤
�𝑗𝑗,𝑡𝑡 is the manager-selected portfolio weight on asset j at the beginning of period

𝑏𝑏
t; 𝑟𝑟̃𝑗𝑗,𝑡𝑡 is the period t return on asset j; and 𝑤𝑤𝑗𝑗,𝑡𝑡 is the benchmark portfolio weight on asset j at the

beginning of time t.

B. Disentangling Manager Actions in Portfolio Performance

Using Equation (2), we can separate the portfolio Active Returns (AR) into Forecasting

Success (FS), Active Management (AM), and Return Dispersion (RD). Because the sum of any

set of portfolio weights equals one, we can add and subtract the cross-sectional mean return, 𝑟𝑟�𝑡𝑡 ,

on the right-hand side of Equation (2) to get

𝑏𝑏 𝑏𝑏
𝐸𝐸�𝑟𝑟̃𝑝𝑝,𝑡𝑡 − 𝑟𝑟̃𝑏𝑏,𝑡𝑡 � = 𝐸𝐸�∑𝑁𝑁 �𝑗𝑗,𝑡𝑡 𝑟𝑟̃𝑗𝑗,𝑡𝑡 − ∑𝑁𝑁
𝑗𝑗=1 𝑤𝑤 𝑗𝑗=1 𝑤𝑤
�𝑗𝑗,𝑡𝑡 𝑟𝑟̃𝑗𝑗,𝑡𝑡 − ∑𝑁𝑁 �𝑗𝑗,𝑡𝑡 𝑟𝑟̅ + ∑𝑁𝑁
𝑗𝑗=1 𝑤𝑤 𝑗𝑗=1 𝑤𝑤
�𝑗𝑗,𝑡𝑡 𝑟𝑟̅ � (3).

Rearranging, we get

6
 𝑏𝑏
𝐸𝐸�𝑟𝑟̃𝑝𝑝,𝑡𝑡 − 𝑟𝑟̃𝑏𝑏,𝑡𝑡 � = 𝐸𝐸�∑𝑁𝑁
𝑗𝑗=1�𝑤𝑤
�𝑗𝑗,𝑡𝑡 − 𝑤𝑤
�𝑗𝑗,𝑡𝑡 � �𝑟𝑟̃𝑗𝑗,𝑡𝑡 − 𝑟𝑟̅ �� (4).

Equation (4) is a modified version of the unconditional portfolio performance measures of Ferson

and Khang (2002) and Grinblatt and Titman (1993). Next, because the cross-sectional mean weight

departures from the benchmark and the cross-sectional mean return differences are both zero, or

���������������
�𝑤𝑤
�𝚥𝚥,𝑡𝑡 − 𝑤𝑤 𝑏𝑏
�𝚥𝚥,𝑡𝑡 � = 0 and ������������
�𝑟𝑟̃𝚥𝚥,𝑡𝑡 − 𝑟𝑟��
𝑡𝑡 = 0, respectively, we can express equation (4) as

𝐸𝐸�𝑟𝑟̃𝑝𝑝,𝑡𝑡 − 𝑟𝑟̃𝑏𝑏,𝑡𝑡 � = 𝐸𝐸 �∑𝑁𝑁

𝑗𝑗=1 ��𝑤𝑤
�𝑗𝑗,𝑡𝑡 − 𝑤𝑤 𝑏𝑏
�𝑗𝑗,𝑡𝑡 � − ���������������
�𝑤𝑤
�𝚥𝚥,𝑡𝑡 − 𝑤𝑤 𝑏𝑏
�𝚥𝚥,𝑡𝑡 ������������
�� ��𝑟𝑟̃𝑗𝑗,𝑡𝑡 − 𝑟𝑟̅ � − �𝑟𝑟̃
𝚥𝚥,𝑡𝑡 − 𝑟𝑟
��𝑡𝑡 �� (5).

Equation (5) shows that in order for a manager to beat the benchmark, there must be an

underlying tendency for the weight departures from the benchmark to covary with subsequent

return differences, or dispersion. In other words, relative to a benchmark portfolio, a skilled

manager will have more money invested in equities when the future return to equities is higher

than to bonds and cash. By assuming a manager’s skill applies equally across all the assets in the

portfolio 5, the right-hand side of Equation (5) becomes N times the covariance between the weight

departures and return differences, i.e.,

E�r�p,t − r�b,t � = NCov�w � tb , 𝑟𝑟̃𝑡𝑡 − 𝑟𝑟��

�t − w 𝑡𝑡 = Nρ�w � tb , 𝑟𝑟̃𝑡𝑡 − 𝑟𝑟��σ�w
�t − w 𝑡𝑡 � tb �σ(𝑟𝑟̃𝑡𝑡 − 𝑟𝑟�)
�t − w 𝑡𝑡 (6).

To simplify the notation, we rewrite our decomposition as

E�r�p,t − r�b,t � = Nρ(∆𝑤𝑤𝑡𝑡 , ∆𝑟𝑟𝑡𝑡 )σ(∆𝑤𝑤𝑡𝑡 )σ(∆𝑟𝑟𝑡𝑡 ) (7a)

AR = N * FS * AM * RD (7b).

Equations (7) contains the terms we define as Active Returns (AR), Forecasting Success

(FS), Active Management (AM), and Return Dispersion (RD). Active Returns (AR) is the average

5
Another approach would be to assume that the covariance is different for each asset, which is similar to Lo (2008).
For example, a manager may know more about the relative returns between equities versus bonds than equities versus
cash. We assume his knowledge is equal across the assets.

7
difference between the portfolio and benchmark returns, E�r�p,t − r�b,t �. Forecasting Success (FS)

is the cross-sectional correlation between the weight departures and subsequent return differences

over the measurement period, ρ(∆𝑤𝑤𝑡𝑡 , ∆𝑟𝑟𝑡𝑡 ). Active Management (AM) is the standard deviation of

the weight departures from the benchmark across assets over the measurement period, σ(∆𝑤𝑤𝑡𝑡 ).

Return Dispersion (RD) is the standard deviation of the return differences from the cross-sectional

mean return over the measurement period, σ(∆𝑟𝑟𝑡𝑡 ). 6

i. Manager Driven Components: Foresight and Active Management

Forecasting Success (FS) is the component of portfolio performance that arises from the

propensity of a portfolio manager to increase (decrease) the weights relative to the benchmark

weights on those assets that subsequently have the higher (lower) returns. Ceteris paribus, the

higher the value of Forecasting Success (FS), the more the original portfolio outperforms the

benchmark. We note that the Forecasting Success (FS) is the component that determines whether

Active Returns (AR) are positive or negative. The other components, which are the number of

securities or are standard deviations, are always positive.

Active Management (AM) is the component of portfolio performance arising from the

aggressiveness a manager uses in choosing portfolio weights that depart from the benchmark

weights. The larger the magnitude of the departures of the portfolio weights from the benchmark

weights, the greater the Active Management (AM). If the manager’s portfolio weights always

equal the benchmark weights, then the standard deviation of the weight departures will be zero. If

6
We note that Lo (2008) on page 7 and a footnote in the CFA Program Curriculum 2017 Level 2 Volume 6 on page
461 by Clark, de Silva, and Thorley (2017) mention a mathematically similar decomposition, but they use it as an
expository tool to explain active management. Here, we use the decomposition as the basis for developing tractable
empirical measures of portfolio performance.

8
there are large departures of the portfolio weights from the benchmark, then the standard deviation

of the weight departures will be large.

Our decomposition shows that Active Management (AM) contributes directly to Active

Returns (AR) by multiplying Forecasting Success (FS). As long as Forecasting Success (FS) is

positive, the higher the Active Management (AM), the greater are Active Returns (AR). Note that

Forecasting Success (FS), however, is not restricted to be positive. A manager can exhibit negative

Forecasting Success (FS), meaning that the correlation between the portfolio weight departures

from the benchmark and subsequent returns is negative. When Forecasting Success (FS) is

negative, the higher the Active Management (AM), the lower are Active Returns (AR), or under-

performance relative to the benchmark.

Finally, we note the relationship between Active Management (AM) measure and the

Active Share (AS) measure of Cremers and Petajisto (2009). Active Management (AM) is the

standard deviation of the weight departures from the benchmark, while Active Share (AS) is N/2

times the mean absolute deviation of the weight departures from the benchmark. 7 So, if the

manager is a “closet indexer,” i.e., holds the benchmark, then both Active Management (AM) and

Active Share (AS) equal zero. The more a manager’s weights depart from the benchmark, the

higher the values for Active Management (AM) and Active Share (AS).

1
7
The Active Share (AS) measure of Cremers and Petajisto (2009) is 𝐴𝐴𝐴𝐴 = ∑𝑁𝑁 𝑏𝑏
�𝑤𝑤 − 𝑤𝑤𝑗𝑗,𝑡𝑡 �. Because the cross-
2 𝑗𝑗=1 𝑗𝑗,𝑡𝑡
sectional mean weight departure ��������������� � − ���������������
1
� ȷ,t − w
�w b
� ȷ,t � = 0, one can write 𝐴𝐴𝐴𝐴 = ∑𝑁𝑁𝑗𝑗=1 ��w� j,t − w b
� j,t �w� ȷ,t − w b
� ȷ,t ��.
2
1
Multiplying AS by 2/N results in the mean absolute deviation of the weight departures, MAD = ∑𝑁𝑁 ���w
� j,t −
𝑁𝑁 𝑗𝑗=1
w b
� j,t � − ���������������
�w� ȷ,t − w b
� ȷ,t ���. Also, note that by Jensen’s Inequality, the Active Management (AM) will always be greater
1
than the mean absolute deviation of the weight departures, because MAD = ∑𝑁𝑁 � j,t − ������
𝑗𝑗=1�∆w ∆w� ȷ,t � =
𝑁𝑁
1 2 1 2
∑𝑁𝑁 � j,t − ������
𝑗𝑗=1 ��∆w ∆w� ȷ,t � ≤ � ∑𝑁𝑁 � j,t − ������
𝑗𝑗=1�∆w ∆w� ȷ,t � = AM.
𝑁𝑁 𝑁𝑁

9
 Both the Active Management (AM) and Active Share (AS) capture the aggressiveness with

which a manager chooses weights that departure from a benchmark. As Cremers and Petajisto

(2009) point out, the appeal of Active Share (AS) is that it captures the amount by which a portfolio

differs from its benchmark. One can interpret Active Share (AS) as representing the percentage of

a portfolio that is longer (or shorter) than the benchmark index. For example, an Active Share (AS)

of 50% means that relative to the benchmark index, the portfolio has 50% more (less) invested in

those stocks where the portfolio exceeds (falls short of) the benchmark weights.

Active Management (AM), on the other hand, is a standard deviation. It is appealing

because Active Management (AM) relates directly to Active Returns (AR), or the returns of a

portfolio in excess of its benchmark, through Equations (7). Finally, Active Management (AM)

has a simple interpretation that makes it appealing when summarizing manager activity. For

b
convenience, assume w
� j,t − w
� j,t is normally distributed. If Active Management (AM) is 5%, then

approximately 68% of the weight departures from the benchmark were within ±5% (i.e., ±1

standard deviation) of zero. In other words, two-thirds of the time the weights for the individual

assets were within ±5% of their benchmark (e.g., if the benchmark equity weight were 50%, then

approximately 68% of the time the equity weight was between 45% and 55%).

ii. Market Driven Component: Return Dispersion

In Equations (7) the component we call Return Dispersion (RD), σ(∆𝑟𝑟𝑡𝑡 ), is the standard

deviation of the return differences from the cross-sectional mean at time t over the measurement

period. Return Dispersion (RD) captures the differences in the returns among a portfolio’s assets,

which make beating or underperforming the benchmark possible. Intuitively, if returns to all the

assets are the same in a given time t, the standard deviation of the return differences will be zero.

10
In this case, all assets have the same return, so there is no Return Dispersion (RD), making RD

equal to zero. This means that it is impossible to beat or underperform the benchmark by choosing

weights that depart from the benchmark weights, since asset returns are all the same. On the other

hand, the higher the Return Dispersion (RD) is across the assets in a portfolio, the greater is the

opportunity to beat the benchmark through the choice of asset weights, since returns among the

assets have greater variation. The investments industry intuitively recognizes this link between

return dispersion and potential performance as is evident in a CNBC article where an analyst states,

“"However, with 2017 shaping up to be a year of higher interest rates and volatility, there is the

potential for higher dispersion amongst stocks and sectors. This represents the last chance for

active managers, who tend to have a quality bias in their asset selection, to try and outperform...." 8

Here, we provide a direct measure of that link and explicitly show how it influences Active Returns

(AR).

Equations (7) also illustrates explicitly a misconception that appears in the popular

financial press. This misconception is the idea that high correlations among assets make it difficult

for active money managers to beat their benchmarks. As stated in a CNBC article, “Beating the

market was especially a challenge as sectors increasingly traded in sync, or with what professional

investors call high correlation. The closer a correlation is to "one," the more likely that moves in

one sector result match moves in another sector or sectors, so it matters less where money is

parked.” 9 However, asset correlation is not a component found in Equations (7). To see why this

is the case, consider two extreme cases. In the first case, suppose correlations are close to one, so

asset returns move together; but Return Dispersion (RD) is very high, so some assets have much

higher returns than others. In this case, managers overweighting the high return assets will

8
http://www.cnbc.com/2016/12/21/2017-is-active-managers-last-chance-to-stop-the-etf-takeover-analyst-says.html
9
http://www.cnbc.com/2017/01/24/why-2017-could-be-the-year-stock-pickers-regain-their-edge.html

11
outperform those managers overweighting the low return assets, regardless of the fact that the asset

returns are moving together. In the second case, suppose correlations are zero, so the assets are

moving independently; but Return Dispersion (RD) is very low. In this case, which assets

managers chose will not matter since asset returns differ only by very small amounts, regardless

of the fact that asset returns are not moving together. In other words, only Return Dispersion (RD)

matters.

Finally, we emphasize that Return Dispersion (RD) is the same for all portfolios using the

same assets over a given measurement period and is outside of a manager’s control. It describes

whether the investment environment was conducive to affecting portfolio returns through altering

portfolio weights. In other words, if Return Dispersion (RD) were higher in period A than in period

B, then the probability of a manager with a high (low) skill level beating (underperforming) the

benchmark would be greater in period A than in period B. Further, cross-sectional differences in

manager Active Returns (AR) will be greater in period A versus period B.

iii. An Illustration of Separating Portfolio Returns into Components

Table 1 illustrates why it is important to separate the effects of Forecasting Success (FS)

and Active Management (AM) on portfolio performance. Consider an example of two managers

who allocate weights among five assets at the beginning of the period as in Table 1. At the end of

the period, both managers beat the benchmark by 5%, as shown by the AR = 0.05 for each manager.

Using Equations (7), Manager A has an FS = 0.9806 and an AM = 0.0894. Manager B has an FS

= 0.2621 and an AM = 0.3347. Because each manager had the same set of assets, RD = 0.1140 for

both managers.

12
 Noting AR = N × FS × AM × RD, Manager A beat the benchmark by 5% because of a

relatively high FS value, meaning that she beat the benchmark by across the board overweighting

the assets that had positive return deviations and underweighting the assets that had negative return

deviations. In contrast, Manager B beat the benchmark by 5% because of a high AM value obtained

through a substantial overweighting in Asset 1 (80% versus 20% in the benchmark), meaning he

beat the benchmark with a big bet on the highest return asset. Manager B incorrectly

underweighted Asset 2. In fact, Manager B shorted this asset even though it had the second highest

positive subsequent return. By knowing the Forecasting Success (FS) and Active Management

(AM) measures, investors will be able to determine whether benchmark beating performance is

driven by allocating weights correctly on most of the assets most of the time or if it is driven by a

few big bets.

IV. Empirical Measures of Active Returns and its Components

A. Specifying the Benchmark Portfolio

To implement our measures for this paper, we need to specify a benchmark. Because we

are examining a set of mutual funds for which we do not have explicitly specified benchmark

portfolios, we follow Ferson and Khang (2002) and Grinblatt and Titman (1993) in using each

fund’s own lagged portfolio weights for the benchmark weights in our empirical analysis. In so

doing, we eliminate the need to come up with an exogenous “pre-specified” benchmark that has a

similar risk profile for each fund and instead derive a benchmark using the weights from each

portfolio itself. For example, specifying a 60% equity, 30% bond, and 10% cash benchmark for

all the funds would be inappropriate. First, given the variation in strategies across funds, it is

unlikely that such a benchmark would be appropriate for each and every fund. Second, if risk

13
varies across assets or is time-varying, then such a fixed benchmark would do a poor job adjusting

for variations in the risk of a portfolio over time. For example, a 60-30-10 benchmark would have

too much risk to be a fair benchmark for a manager whose task is to manage a low risk portfolio

where a large portion of the assets stay in cash.

Using the lagged weights provides a passive benchmark that effectively judges the manager

against a buy-and-hold portfolio in each period. 10 It is important to note that the buy-and-hold

adjusts over time so the benchmark will track the measured portfolio’s risk with a one period lag.

Another way to think of this is that the weight changes from the previous period are the relevant

Active Management (AM) decision. Of course, one can use an investor/client–assigned

benchmark, if an explicit one exists. However, in lieu of explicit benchmarks, the lagged weight

benchmark is the most practical method to use when examining the performance of a large sample

of funds for which explicit benchmarks are unavailable.

Just as important, using a lagged weight benchmark mitigates the need to specify an asset

pricing model to adjust for differences in the risks and expected returns of the individual assets.

For example, stocks have higher expected returns and risk than cash. To the extent that a portfolio

manager maintains a higher weight in equities than cash relative to the benchmark, we would

expect the manager to outperform the benchmark in terms of returns (as well as have greater risk).

However, using the lagged weights of the portfolio as the benchmark mitigates this issue, because

the weights of the benchmark will, most of the time, be similar to the actual portfolio. Thus, the

portfolio and the benchmark will have similar expected return and risk, with the Active Return

(AR) capturing any marginal additional return obtained by changing the portfolio weights. We

10
Technically, a buy-and-hold portfolio would update the lagged weights at the beginning of each period by the returns
over the period (See Ferson and Khang (2002). Here, we follow Grinblatt and Titman (1993) and use the unadjusted
lagged weights as a simplified version of buy-and-hold.

14
note that for robustness, we performed our analyses using abnormal returns calculated by

subtracting the mean return for each asset over the measurement period and obtained nearly

identical results. This confirms that using the lagged weight benchmark creates a benchmark

portfolio with similar risk and expected return to the original portfolio for the funds in our sample.

B. Empirically Measuring Active Returns and its Components

Our empirical measures of Forecasting Success (FS), Active Management (AM), and

Return Dispersion (RD) follow directly from Equations (7). Our method for estimating these

components recognizes that the excess return to beating a benchmark at any time t (ARt) is the

multiplicative product of the three individual return components (FSt, AMt, RDt). Thus, we cannot

simply take averages of each return component over time. Instead, we must calculate the statistics

jointly over the entire measurement period. By doing this, we preserve the multiplicative aspect of

the relationship in Equations (7) that exists between the excess return over the benchmark and the

individual return components.

Equations (8) below contain the corresponding empirical measures we use to calculate

Active Returns (AR), Forecasting Success (FS), Active Management (AM), and Return Dispersion

(RD). These empirical measures are values over the entire measurement period for a given fund.

Thus,

𝐴𝐴𝐴𝐴 = ������������
𝑟𝑟𝑝𝑝,𝑡𝑡 − 𝑟𝑟𝑏𝑏,𝑡𝑡 ∀𝑡𝑡 (8a)

FS = ρ�w
� j,t − w
� j,t−1 , 𝑟𝑟̃𝑗𝑗,𝑡𝑡 − 𝑟𝑟��
𝑡𝑡 ∀𝑗𝑗, 𝑡𝑡 (8b)

AM = σ�w
� j,t − w
� j,t−1 � ∀𝑗𝑗, 𝑡𝑡 (8c)

RD = σ�𝑟𝑟̃𝑗𝑗,𝑡𝑡 − 𝑟𝑟��
𝑡𝑡 ∀𝑗𝑗, 𝑡𝑡 (8d),

15
where j is either equities, bonds, or cash and t is time. Note the Return Dispersion (RD) measure

will be the same for all funds measured over the same time period. The Forecasting Success (FS)

and Active Management (AM) measures, however, are specific to a given portfolio over a given

time period.

Finally, we note that Equations (8) are empirical measures implementing Equations (7), so

it is an empirical question as to whether the equality in Equation (7b) holds using our estimation

method. Verifying these predictions is important because Equations (7) hold exactly at each time

t, but our estimation method provides estimated values across all t. We verify that 𝐴𝐴𝐴𝐴 = 𝑁𝑁 ∗ 𝐹𝐹𝐹𝐹 ∗

𝐴𝐴𝐴𝐴 ∗ 𝑅𝑅𝑅𝑅 is true across the sample. 11

V. Empirical Results

A. Measuring Active Returns

We use is a sample of 732 US asset allocation mutual funds from Morningstar. Each fund

is in one of the Morningstar categories of Aggressive Allocation, Moderate Allocation, and

Conservative Allocation. These categories represent funds that invest primarily in equities, bonds,

and cash. The earliest fund observation is December 1983, while the latest observation is June

2013. Among the funds, the median first year observation is in 2002, while the median last year

observation is in 2013. Further, 75% of the funds have a first year observation during or after 1997,

meaning most of the fund observations occur between 1997 and 2013. The reporting frequency for

11
Results available upon request. We linearize the functional form in Equations (7) by taking the natural log of both
sides to transform the multiplicative relationship into a linear one. We use the absolute value of the Active Returns
(AR) and the Performance Coefficient (PC) when their values are negative. The estimated equation is ln(|𝐴𝐴𝐴𝐴|)𝑚𝑚 =
𝛾𝛾0 + 𝛾𝛾𝑃𝑃𝑃𝑃 ln(|𝑃𝑃𝑃𝑃|)𝑚𝑚 + 𝛾𝛾𝐴𝐴𝐴𝐴𝐴𝐴 ln(𝐴𝐴𝐴𝐴𝐴𝐴)𝑚𝑚 + 𝛾𝛾𝑅𝑅𝑅𝑅𝑅𝑅 ln(𝑅𝑅𝑅𝑅𝑅𝑅)𝑚𝑚 + 𝜖𝜖𝑚𝑚 , where m represents the number of mutual funds.
Note that the natural log is undefined for values less than or equal to zero, and none of the funds had performance
equal to exactly zero. The predicted coefficients are 𝛾𝛾0 = 1.0986 [= ln(3) because N=3 assets in our case] and 𝛾𝛾𝑃𝑃𝑃𝑃 =
𝛾𝛾𝐴𝐴𝐴𝐴𝐴𝐴 = 𝛾𝛾𝑅𝑅𝑅𝑅𝑅𝑅 = 1. Finally, we compute the empirical version of Equation (7b) for each fund and verify the
relationship.

16
most of the funds is either monthly, quarterly, or semi-annually. Some funds had a mixture of

these.

To proxy for equity, bond, and cash returns, we use the returns to the S&P 500, the US 10-

year Treasury Bond, and LIBOR, respectively. We believe these return series capture the broad

choice between equities, bonds, and cash, which is the primary decision for these asset allocation

funds. However, if we wished to capture departures from these return series, we could use

abnormal returns from an asset pricing model (eg. Fama-French to capture a small cap tilt in the

equity allocation), but this would pull the focus away from examining the asset allocation decision

directly.

As stated in Section IV, the benchmark in our empirical implementation here is the lagged

𝑏𝑏
benchmark weights. Hence, 𝑤𝑤
�𝑗𝑗,𝑡𝑡 = 𝑤𝑤
�𝑗𝑗,𝑡𝑡−1 , making Equation (4)

𝐴𝐴𝐴𝐴 = 𝐸𝐸�∑𝑁𝑁
𝑗𝑗=1�𝑤𝑤
�𝑗𝑗,𝑡𝑡 − 𝑤𝑤
�𝑗𝑗,𝑡𝑡−1 � �𝑟𝑟̃𝑗𝑗,𝑡𝑡 − 𝑟𝑟̅ �� (9).

Here the lagged weights are the last reported weights. We can think of departures from the most

recently reported weights as departures from a buy-and-hold portfolio in each period, so the Active

Management decision is whether to change the weights from what they were previously. Further,

using the lagged weights creates a benchmark portfolio with similar risk and expected return to

the original portfolio.

If a fund has monthly observations, then we simply use those to create the monthly weight

changes. If a fund has a longer sampling period, then we fill in the months with the most recently

observed weight change. This sampling method provides us with monthly changes that account

for the apparent length of time that the manager appears to have committed to having the new asset

allocation in place.

17
 To illustrate, consider a couple of examples where it is currently the end of December and

a manager’s asset allocation to equities, bonds, and cash at the end of November, the prior month,

was (60%, 30%, 10%). In the first example, the manager believes stock returns will be higher in

January, so he changes his asset allocation to (70%, 25%, 5%) at the end of December. This means

the weight changes used to estimate Active Returns (AR), or the returns in excess of the single

period buy-and-hold benchmark, for January will be (10%, -5%, -5%). Note these weight changes

are relative to a single period buy-and-hold benchmark of (60%. 30%, 10%). To get the Active

Returns (AR), we multiply the weight changes by the return differences in the month of January.

If the manager changes his weights back to (60%, 30%, 10%) by the end of January, then we know

that his weights at the end of December, the prior month, were based on an expectation of a relative

return advantage for equities in January that no longer exists by the end of January. The single

period buy-and-hold benchmark would now be (70%. 25%, 5%). The weight changes of (-10%,

5%, 5%) would then be multiplied by February’s returns to get Active Returns (AR).

Now suppose the manager does not change his weights back to (60%, 30%, 10%) until the

end of March because he believes equities will have a relative return advantage over the next three

months. In this case the weight changes would be (10%, -5%, -5%) for January, February, and

March. In this way, we account for the fact that the manager has a higher allocation to equities for

the entire three-month period, based on his decision not to change his asset allocation back until

the end of March. Thus, our methodology assumes that each observation is a decision by the

manager to have the asset allocation observed rather than the previous asset allocation, regardless

of how long ago the previous allocation occurred.

Finally, note that if instead of decreasing his equity allocation, the manager further

increased his allocation to equities at the end of March to (75%, 20%, 5%), then this interpretation

18
does not change. It simply indicates a belief that the relative return advantage of equities has

become even greater than before and his weight changes will now reflect this, making the new

weight changes (5%, -5%, 0%).

Lastly, to strike a balance between survivorship bias and having an adequate number of

observations to measure fund performance, we kept all funds that had at least six separate reported

observations. This means all funds in the sample have data spanning a minimum of six months.

We also eliminate any fund that had any consecutive reported observations with more than two

years between them. This gives us a sample of 732 separate mutual funds with 46,609 reported

fund-asset allocation observations and 91,10 fund-month observations, indicating the average time

between reported fund observations is about two months.

B. Sample Statistics for the Funds

Table 2 reports fund characteristics for the 732 mutual funds in the sample and the

Morningstar categories of Aggressive Allocation (n=112), Moderate Allocation (n=383), and

Conservative Allocation (n=237). Panel A shows the distribution of fund percentage weight

allocations for the entire sample and Morningstar categories. The average weights for the entire

sample are 54.76% in equities, 34.16% in bonds, and 11.07% in cash. However, there is a

significant distribution across funds. For example, the average stock allocation ranged from

13.56% at the 2.5th percentile to 85.14% at the 97.5th percentile.

The differences between Aggressive and Conservative funds is evident in the distributions

of allocations, as well. Aggressive funds had much higher average equity allocations of 77.48%

with lower bond and cash weights of 15.04% and 7.48%, respectively. Conservative funds had

lower average equity allocations of 34.46% with higher bond and cash weights of 50.61% and

19
14.93%, respectively. Moderate funds had weights similar to the average for the entire sample at

60.68% equities, 29.58% bonds, and 9.74% cash. The wide variation in the distribution of weights

indicated by these figures is strong evidence in favor of our use of the lagged buy-and-hold weights

to construct our benchmarks, as this will create benchmarks that track both cross-sectional and

time series variation in fund risk.

Panel B reports on the distribution of fund weight changes. For the median fund, the mean

absolute deviation (MADW) is 4.48% for equity changes, 4.08% for bond changes, and 3.86% for

cash changes. This result suggests that the typical weight change for the average fund is smaller

than 5% (e.g. a move smaller than one from 50% equity to 55% equity).

Panel B also reports the Active Share (AS) measure of Cremers and Petajisto (2009) and

our Active Management (AM) measure. Note that Active Share (AS) is simply N/2 (or 3/2 in this

case) times the MADW, so the two measures differ cardinally by a fixed multiple, but are ordinally

identical. Also, AM will always exceed MADW by Jensen’s Inequality (see footnote 7).

Interestingly, as one moves across the Morningstar categories of Aggressive, Moderate, and

Conservative, the average values for the MADW, Active Share (AS), and the Active Management

(AM) increase. This relationship may seem counterintuitive, but emphasizes that the Morningstar

categories differentiate by the amount of equity investment versus the less volatile assets of bonds

and cash and not by the degree of activity level in changing the portfolio weights. “Aggressive”

by Morningstar’s definition refers to the relative investment in the riskier asset (in this case,

equity), and not to whether the manager aggressively manages the portfolio by making large and

frequent portfolio weight changes. Interestingly, the funds in the Conservative category appear to

be the most active in managing their portfolio weights.

20
 Panel C shows the distribution of the mean absolute deviation of the return differences

(MADR), average raw return differences across the various fund measurement periods, and Return

Dispersion (RD). The MADR indicates that time t stock and bond returns differ from the time t

mean return by about plus or minus 2.77% and 2.69%, respectively. Time t cash returns tended to

be about 50% closer to the time t mean return, since its average MADR is only 1.35% by

comparison. The average raw return differences indicate that both time t stock and bond returns

averaged 0.17% above the time t mean return, while time t cash returns averaged 0.35% below the

time t mean return. That stock and bond returns were higher than cash returns on average is not

surprising. That stock and bond returns had similar averages is an artifact of the sample period,

because most of the fund-observations span the period from 1997 to 2013.

Panel D reports various sample statistics. The first is the number of reported asset

allocation observations by each fund. For most funds, these occur either monthly or quarterly. The

average is 60 reported observations per fund. The second is the number of months over which

fund performance is measured. The average measurement period is 120 months. Not surprisingly,

the average time between observations is two months. Note that these three statistics are consistent

with about half the funds having monthly observations and the other half having quarterly

observations. Finally, Panel D reports the distribution of fund size in terms of the average dollars

invested over the measurement period. The median fund size was $115.8 million.

C. Active Returns and its Components for the Fund Sample

Table 3 reports statistics for Active Returns (AR), Forecasting Success (FS), Active

Management (AM), and the Return Dispersion (RD) for the 732 mutual funds in our sample. Panel

A reports the distributions for AR, FS, AM, and RD. The table reports these for the sample as a

21
whole, and the three Morningstar categories of Aggressive, Moderate, and Conservative.

Interestingly, we observe few differences from the sample as a whole versus the Morningstar

categories.

Examining the Active Returns (AR) for the sample as a whole, it is apparent that the

average mutual fund does not outperform the lagged weight benchmark. The average and median

AR is close to zero. The distribution of the AR t-statistics, however, indicates that some funds may

outperform the benchmark. While the distribution of the average Active Returns (AR) centers

about zero, the t-statistic at the 95th percentile is 1.80, which is consistent with significance at the

10% level; and the t-statistic at the 97.5th percentile is 2.02, which is consistent with significance

at the 5% level. On the flipside, there are funds in the sample that significantly underperform their

benchmarks at similar levels at the 2.5th and 5th percentiles.

These results are consistent with one of two interpretations. The first is that the sample

shows that managers have no skill, and the performance at the tails of the distribution occur by

chance. The second is that the average manager has no skill, but those at the tails of the distribution

do have differentiated skill that is captured by the distribution. Differentiating between these two

interpretations requires an examination of performance persistence. Later, we examine whether

average Active Returns (AR) and its components persist from one period to the next. If the

interpretation of luck is correct, then we should find no evidence of persistence. If the interpretation

of skill is correct, then we should find persistence in Active Returns (AR) and Forecasting Success

(FS).

The Forecasting Success (FS) distributions are consistent with the Active Returns (AR)

results. The distribution of Forecasting Success (FS) centers near zero. In the upper tail, however,

the Forecasting Success (FS) is a positive 0.1546 at the 95% percentile and 0.2072 at the 97.5th

22
percentile. The lower tail 2.5th and 5th percentiles indicate similar magnitude Forecasting Success

(FS), but with the opposite sign.

The Active Management (AM) measure is consistent with about 68% of the weight changes

being within plus or minus 7.92% (assuming a normal distribution). Thus, most of the fund weight

changes appear to be less than 10%, meaning most of the weight changes are not too dramatic.12

For example, consider an initial stock allocation of 60%. A move to a 68% or a 52% stock

allocation would be larger than about 68% of those observed.

The Return Dispersion (RD) measure indicates that about 68% of the return differences are

within plus or minus 3.08% on a monthly basis. In other words, the average difference between

equities, bonds, and cash versus the mean return in a given month t tends to be less than 3% about

68% of the time (assuming a normal distribution). For example, consider a stock return of 6%, a

bond return of 3%, and a cash return of 0%. The average return across assets would be 3%,

providing a differential for equities of 3%, a differential for bonds of 0%, and a differential for

cash of -3%. About 68% of the months would have smaller differences than these. Note that the

greater the Return Dispersion (RD), the greater the potential for variations in asset allocation to

generate positive or negative Active Returns (AR).

Across the Morningstar categories, the Active Returns (AR) and its components are

similar. The distribution of Active Returns (AR) and Forecasting Success (FS) center about zero

for each category and the sample as a whole. The Return Dispersion (RD) is just over 3% for each

category, which is consistent with the funds in each category having similar measurement periods.

The one notable difference across categories is that the average (median) Active Management

(AM) for the Conservative category at 8.52% (6.73%) is higher than for the Moderate and

12
We note that internal fund guidelines or regulations likely influence individual fund AMs.

23
Aggressive categories at 7.90% (5.81%) and 6.73% (3.91%), respectively. Thus, while the

Conservative funds are more conservative in terms of the risk of their underlying assets,

Conservative funds are more active in varying their portfolio weights.

Panel B shows that the correlations between the Active Management (AM) versus the

MADW and Active Share (AS) are 0.8839 and above. 13 The results show that Active Management

(AM), MADW, and Active Share (AS) arrive at similar conclusions when evaluating the levels of

Active Management in funds. Panel B also shows the correlations between Return Dispersion (RD)

and MADR. These two measures of cross-sectional return variation are also highly correlated at

0.9771 and above. As discussed earlier, the advantage of the Active Management (AM) and Return

Dispersion (RD) is that they are directly related to Active Returns (AR), which makes clear the

role that each plays in influencing whether a portfolio beats its benchmark or not.

D. Making Comparisons between Selected Funds

Table 4 reports comparisons of a few selected funds to illustrate the information contained

in Forecasting Success (FS), Active Management (AM), and Return Dispersion (RD). 14 Panel A

presents comparisons isolating the effect of Forecasting Success (FS) on Active Returns (AR),

where we match funds that have similar Active Management (AM) and Return Dispersion (RD).

Panel B presents comparisons isolating the effect of Active Management (AM) on Active Returns

(AR), where we match funds that have similar Forecasting Success (FS) and Return Dispersion

(RD). Panel C presents comparisons isolating the effect Return Dispersion (RD) on Active Returns

(AR), where we match funds that have similar Forecasting Success (FS) and Active Management

(AM). Note that each Panel has two cases, one where Forecasting Success (FS) is positive, and

13
Active Share (AS) and MADW have a correlation of one by definition, so they are not in shown (see footnote 7).
14
The actual identities of the funds are unimportant for our purposes, but available upon request.

24
one where it is negative. This emphasizes that Forecasting Success (FS) determines whether Active

Returns (AR) are positive or negative.

The first comparison in Panel A shows the effect of higher Forecasting Success (FS) when

Forecasting Success (FS) is positive, meaning that both funds beat the benchmark. Fund A1 has

an annualized AR that is 3.33% higher than Fund A2. This is due primarily to Fund A1’s

Forecasting Success (FS) being higher than Fund A2’s (0.2210 vs 0.0490). While a part of this

difference is also due to Fund A1’s Active Management (AM) and Return Dispersion (RD) being

a bit higher, these differences are small.

The second comparison in Panel A shows the effect when a fund has a more negative

Forecasting Success (FS). Fund A3 has an annualized AR that is 4.18% lower than Fund A4. Fund

A3’s more negative AR is due primarily to its more negative Forecasting Success (FS) compared

to that of Fund A4’s (-0.1577 vs -0.0563). Note that a small part of this difference is due to Fund

A3 having a higher Return Dispersion (RD), though the lower Active Management (AM) cancels

some of the higher Return Disperion (RD) effect.

The first comparison in Panel B shows the effect of higher Active Management (AM) when

Forecasting Success (FS) is positive. Fund B1 has an annualized AR that is 3.15% higher than

Fund B2. This is due primarily to Fund B1’s Active Management (AM) being higher than Fund

B2’s (23.47% vs 2.64%), which magnifies Fund B2’s Forecasting Success (FS). Note that Fund

B1’s Forecasting Success (FS) and Return Dispersion (RD) are slightly lower or the difference in

annualized AR would be slightly greater.

The second comparison in Panel B shows the effect of higher Active Management (AM)

when Forecasting Success (FS) is negative. Fund B3 has an annualized AR that is 5.29% lower

than Fund B4. This is due to Fund A3’s higher Active Management (AM) versus Fund A4’s

25
(31.60% vs 3.91%), which magnifies Fund A3’s negative Forecasting Success (FS). Note Fund

A3 and A4 have similar Forecasting Success (FS) and Return Dispersion (RD).

The first comparison in Panel C shows the effect of higher Return Dispersion (RD) when

Forecasting Success (FS) is positive. Fund C1 has an annualized AR that is 0.11% higher than

Fund C2. This is due to Fund C1’s Return Dispersion (RD) being higher than Fund C2’s (3.82%

vs 3.18%), which magnifies its Forecasting Success (FS). Note that Fund C1 and C2 have similar

Forecasting Success (FS) and Active Management (AM). Also, note that the relatively small

difference in annualized AR is due to a lack of variation in Return Dispersion (RD) (as compared

to variation in Active Management (AM)). This is true for much of the sample because the

measurement periods for most funds overlap.

The second comparison in Panel C shows the effect of higher Return Dispersion (RD) when

Forecasting Success (FS) is negative. Fund C3 has an annualized AR that is 0.27% lower than

Fund C4. This is due to Fund C3’s Return Dispersion (RD) being higher than Fund C4’s (2.99%

vs 2.12%), which magnifies its negative Forecasting Success (FS). Note that absolute magnitude

of Fund C3’s Forecasting Success (FS) and Active Management (AM) are slightly lower or the

difference in annualized AR would be slightly greater.

E. Plotting Fund Active Returns and Manager Driven Components

In Figure 1, we show a plot of our sample of 732 funds with Forecasting Success (FS) on

the x-axis and Active Returns (AR) on the y-axis. This plot shows the positive relation between

Forecasting Success (FS) and Active Returns (AR). Figure 2 is a plot of Active Management (AM)

on the x-axis and Active Returns (AR) on the y-axis. This plot illustrates that greater Active

Management (AM) leads to a greater dispersion in Active Returns (AR). Figure 3 is a plot of

26
Forecasting Success (FS) on the x-axis and Active Management (AM) on the y-axis. This plot

shows that Forecasting Success (FS) seems directionally unrelated to Active Management (AM).

However, there is a tendency for funds with the most extreme Active Management (AM) to have

Forecasting Success (FS) closer to zero.

To illustrate the information in the plots, we also highlight the two extreme funds in terms

of Active Returns (AR). Fund A has the highest Active Returns (AR) of the funds in the sample.

The high Active Returns (AR) are the result of having the third highest Forecasting Success (FS)

combined with one of the higher results for Active Management (AM). Fund B has the lowest

Active Returns (AR) among the funds in the sample. The low Active Returns (AR) are the result

of having one of the lower Forecasting Success (FS) results combined with one of the higher Active

Management (AM) results.

F. Persistence of Active Returns and its Components

Table 5 examines the predictability and persistence of those elements of a fund’s

performance that are under its control. This will provide evidence on whether the funds at the tails

of the distributions of Active Returns (AR) , Forecasting Success (FS), and Active Management

(AM) are there because of differences in skill level or by chance. If there is skill, we would expect

to observe evidence of persistence. If by chance, then we should observe no persistence.

To examine persistence, we divide the observations for each fund into a first half and

second half. We eliminate any fund that has fewer than 26 reported observations, meaning it must

have at least 13 in each subsample. This reduces the number of funds to 678 from 732. We then

examine three questions. First, do Active Returns (AR) persist from the first half to the second half

subsample. Second, does knowing about Forecasting Success (FS1), Active Management (AM1),

27
and Return Dispersion (RD1) in first half subsample give us any information about Active Returns

(AR2) in the second half subsample. Finally, does Forecasting Success (FS) and Active

Management (AM) persist from the first half to the second half subsample.

The first column in Table 5 presents the results of a regression of Active Returns (AR2)

from the second subsample on past Active Returns (AR1) in the first subsample. There appears to

be no evidence that subsequent Active Returns (AR2) are related to past Active Returns (AR1).

This result is consistent with the Active Returns (AR) distribution being due largely to chance and

is consistent with Fama and French (2010), who argue that the distributions of fund returns are

generated by chance.

The second column of Table 5 shows a regression of Active Returns (AR2) on past

Forecasting Success (FS1), Active Management (AM1), and Return Dispersion (RD1). Consistent

with the result in the first column, there is no evidence that any of these components from the past

provide information about future Active Returns (AR2).

The third column of Table 5 presents the results of a regression of Forecasting Success

(FS2) on past Forecasting Success (FS1), Active Management (AM1), and Return Dispersion

(RD1). Interestingly, this regression indicates that Forecasting Success (FS) from the first half

subsample (FS1) is positively related to subsequent Forecasting Success (FS2) in the second half

subsample. This indicates that past Forecasting Success (FS1), or past success in prediction, has

some tendency to persist over time. However, this does not translate into persistence in Active

Returns (AR). This may be due to a negative relation between past Forecasting Success (FS1) and

subsequent Active Management (AM2), which we discuss below.

In column four, Table 5 shows the results of a regression of Active Management (AM2)

on past Forecasting Success (FS1), Active Management (AM1), and Return Dispersion (RD1).

28
Here, there is strong evidence of persistence as Active Management (AM2) is strongly positively

related to past Active Management (AM1). This indicates that funds who have aggressively altered

their portfolios in the past, resulting in high Active Management (AM1), have a strong tendency

to continue that approach in the future. This persistence in Active Management (AM) is important

because funds with high Active Management (AM) will tend to either out- or under-perform by

more than those with lower Active Management (AM). This is true because Active Management

(AM) multiplies Forecasting Success (FS) to affect Active Returns (AR). Interestingly, Active

Management (AM2) is also negatively related to past Foresight (FS1). This may why Active

Returns (AR) do not persist. Positive Forecasting Success (FS1) may subsequently lead to less

Active Management (AM2), and/or Negative Forecasting Success (FS1) may subsequently lead to

more Active Management (AM2).

To examine this question more closely, we add a dummy variable, which is equal to 1 if

past Forecasting Success (FS1) is negative and 0 otherwise. We also interact this dummy variable

with past Forecasting Success (FS1). Column five of Table 5 presents these results. They indicate

that the negative relation between Active Management (AM2) and past Foresight (FS1) is

restricted to those cases where Foresight (FS1) is negative. This is consistent with negative

Forecasting Success (FS1) subsequently leading to more Active Management (AM2). Thus,

managers who experience past failure in prediction along with the related negative Active Returns

(AR1) become more aggressive in the subsequent period, perhaps in an attempt to make up for the

associated past negative Active Returns (AR1).

29
VI. Conclusion

Measuring portfolio performance has long been a focus of financial economists. Generally

scarce in the extensive literature on measuring portfolio performance are studies that directly

examine the actions of active portfolio managers. Wermers (2003), Lo (2008), and Cremers and

Petajisto (2009) are the first empirical studies to attempt to measure and relate active management

to performance.

In this paper, we add to this literature by starting with portfolio returns and applying a

simple decomposition that intuitively shows how a manager’s actions affect Active Returns (AR),

or returns in excess of a benchmark. In doing this, we provide insight into the performance of

portfolios that goes beyond simply whether they beat a benchmark.

Using a sample of asset allocation mutual funds from Morningstar, we illustrate how

Forecasting Success (FS), Active Management (AM), and Return Dispersion (RD) can explain

differences in fund Active Returns (AR). We show that Forecasting Success (FS) determines

whether Active Returns (AR) are positive or negative, and that Active Management (AM)

magnifies how large those positive or negative Active Returns (AR) will be.

We find that the average and median Active Returns (AR) across funds is close to zero,

implying that the average mutual fund does not outperform a lagged buy-and-hold benchmark.

While some of the funds at the tails of the distribution do outperform the benchmark, we find a

lack of persistence in Active Returns (AR), which indicates that this outperformance is random

and due to chance. This is consistent with the results of Fama and French (2010).

In examining the components of Active Returns (AR) that are under a funds control, we

find weak persistence in Forecasting Success (FS) and strong persistence in Active Management

(AM). The weak persistence in Forecasting Success (FS) suggests that success in predicting past

30
returns has some tendency to persist. However, this does not translate into persistence in Active

Returns (AR), perhaps due to a negative relationship between past Forecasting Success (FS1) and

subsequent Active Management (AM2). The strong persistence in Active Management (AM)

indicates that funds stick to the same level of activity over time in terms of altering their portfolio

weights. This strong persistence in the Active Management (AM) is important because funds with

high Active Management (AM) will have more extreme (both good and bad) performance than

funds with lower Active Management (AM). Finally, we find evidence that negative past

Forecasting Success (FS1) and the associated negative past Active Returns (AR1) leads to greater

subsequent Active Management (AM2), indicating that funds respond to poor past performance

by making larger bets in an attempt to overcome that poor past performance.

31
References

Bange, M., Miller, T., Jr. 2004. Return momentum and global portfolio allocations. Journal of
Empirical Finance 11, 429–459.
Bange, M., Khang, K., Miller, T., Jr. 2008. Benchmarking the performance of recommended
allocations to equities, bonds, and cash by international investment houses. Journal of
Empirical Finance 15, 363–386.
Barras, L., Scaillet, O., Wermers, R. 2010. False discoveries in mutual fund performance:
Measuring luck in estimated alphas. Journal of Finance 65, 179–216.
Carhart, M. 1997. On persistence in mutual fund performance. Journal of Finance 52, 57–82.
Cheng, Evelyn (2017, January 24). Why 2017 could be the year stock pickers regain their edge.
CNBC. Retrieved from http://www.cnbc.com/2017/01/24/why-2017-could-be-the-year-
stock-pickers-regain-their-edge.html.
Copeland, T., Mayers, D. 1982. The Value-Line enigma (1965-1978): A case study of performance
evaluation issues, Journal of Financial Economics 10, 289–322.
Cremers, M., Petajisto, A. 2009. How active is your fund manager? A new measure that predicts
performance. Review of Financial Studies 22, 3329–3365.
Daniel, K., Titman, S. 1997. Evidence on the characteristics of cross-sectional variation in
common stock returns. Journal of Finance 52, 1–33.
Fama, Eugene F., and Kenneth R. French. 2010. “Luck versus Skill in the Cross-Section of Mutual
Fund Returns.” Journal of Finance, vol. 65. no. 5 (October): 1915-1947.
Ferson, W., Khang, K. 2002. Conditional performance measurement using portfolio weights:
Evidence for pension funds. Journal of Financial Economics 65, 249–282.
Grinblatt, M., Titman, S. 1993. Performance measurement without benchmarks: An examination
of mutual fund returns. Journal of Business 60, 97–112.
ICI 2016 Investment Company Fact Book. Retrieved from www.ici.org
Kacperczyk, M., Sialm, C., Zheng, L. 2005. On the industry concentration of actively managed
equity mutual funds. Journal of Finance 60, 1983–2011.
Lo, A. H. 2008. Where Do alphas come from?: A new measure of the value of active investment
management. Journal of Investment Management 6, 1-29.

32
Melloy, John (2016, December 21). 2017 is active managers' 'last chance' to stop the ETF takeover,
analyst says. CNBC. Retrieved from http://www.cnbc.com/2016/12/21/2017-is-active-
managers-last-chance-to-stop-the-etf-takeover-analyst-says.html.
Rekenthaler, John (2014, May 29). Active Share: What You Need to Know. Morningstar.
Retrieved from http://www.morningstar.com/advisor/t/92563289/active-share-what-you-
need-to-know.htm.
Roll, R. 1992. A mean/variance analysis of tracking error. Journal of Portfolio Management 18,
13-23.
Wermers, R. 2006. Performance evaluation with portfolio holdings information. North American
Journal of Economics and Finance 17, 207–230.
Wermers, R. 2003. Are mutual fund shareholders compensated for active management “Bets”?
Working paper, University of Maryland.
Wermers, R. 2000. Mutual fund performance: An empirical decomposition into stock-picking
talent, style, transaction costs, and expenses. Journal of Finance 55, 1655–1703.

33
 Table 1

Illustrating the Importance of Decomposing Performance

Decomposing Portfolio Returns of Two Hypothetical Managers

Asset 1 Asset 2 Asset 3 Asset 4 Asset 5

Benchmark Weights 0.20 0.20 0.20 0.20 0.20

Manager A Weights 0.30 0.30 0.20 0.10 0.10

Manager A Active Weights 𝑤𝑤 �𝑡𝑡𝑏𝑏
�𝑡𝑡 − 𝑤𝑤 0.10 0.10 0.00 -0.10 -0.10

Manager B Weights 0.80 -0.20 0.00 0.20 0.20

Manager B Active Weights 0.60 -0.40 -0.20 0.00 0.00

Asset Returns 0.15 0.10 0.00 -0.10 -0.15

Return Differences 𝑟𝑟̃𝑡𝑡 − 𝑟𝑟̅𝑡𝑡 0.15 0.10 0.00 -0.10 -0.15

AR FS AM RD
Manager A 0.05 0.9806 0.0894 0.114
Manager B 0.05 0.2621 0.3347 0.114
 Table 2

Characteristics of Funds in Sample

Table 2 reports on mutual fund characteristics for the 732 mutual funds in the sample. Panel A shows the distribution
of fund percentage weight allocations for the sample as well as three of the Morningstar categories. Panel B reports
the distribution of fund weight departures (or changes since using a lagged weight benchmark) for the entire sample
and three Morningstar categories. The statistics reported are Active Management (AM), mean absolute deviation of
the weight changes (MADW) and Active Share (AS) (see Cremers and Petajisto (2009)). Note that the AS is simply
3/2 times MADW. Panel C shows the distribution of the mean absolute deviation of return differences (MADR) and
raw return differences across the measurement periods of the funds. Panel D reports on the distribution of the number
of reported asset allocation observations, the distribution of the number of months in each fund measurement period,
the distribution of time between reported asset allocation observations, and the distribution of average size of the funds
in millions of dollars.

Sample Sample Percentiles

Sample Fund Characteristics (M=732) Mean Median 2.5% 5.0% 95.0% 97.5%

Panel A
All Funds (M=732)
Average Allocation to:
Equities 54.76% 58.41% 13.56% 19.42% 81.59% 85.14%
Bonds 34.16% 33.05% 5.76% 8.38% 63.34% 68.68%
Cash 11.07% 7.76% 1.56% 2.00% 31.49% 40.31%
Aggressive Sample (M=112)
Average Allocation to:
Equities 77.48% 77.69% 60.70% 65.92% 89.94% 93.60%
Bonds 15.04% 15.20% 1.14% 2.78% 27.36% 28.84%
Cash 7.48% 5.94% 1.68% 1.94% 17.65% 26.48%
Moderate Sample (M=383)
Average Allocation to:
Equities 60.68% 60.76% 38.26% 47.08% 74.45% 77.55%
Bonds 29.58% 30.67% 7.83% 10.76% 41.79% 44.56%
Cash 9.74% 6.66% 1.61% 1.98% 28.34% 41.00%
Conservative Sample (M=237)
Average Allocation to:
Equities 34.46% 34.70% 6.61% 10.56% 56.36% 62.08%
Bonds 50.61% 51.49% 17.30% 26.33% 71.23% 75.90%
Cash 14.93% 11.61% 0.89% 2.46% 36.65% 42.40%

Panel B
Sample (M=732)
Mean absolute deviation of weight changes for:
Equities 4.48% 3.26% 0.77% 0.92% 11.92% 16.68%
Bonds 4.08% 3.14% 0.68% 0.84% 9.75% 13.67%
Cash 3.86% 2.56% 0.57% 0.74% 10.99% 16.47%

Sample (M=732)
Mean absolute deviation of weight changes across all assets 4.14% 3.14% 0.80% 0.98% 10.26% 14.94%
Active Share (AS) 6.21% 4.71% 1.21% 1.47% 15.39% 22.40%
AM 7.92% 5.83% 1.19% 1.46% 21.53% 27.03%
Aggressive Sample (M=112)
Mean absolute deviation of weight changes across all assets 3.13% 2.35% 0.64% 0.71% 8.96% 10.24%
Active Share (AS) 4.70% 3.52% 0.96% 1.07% 13.45% 15.36%
AM 6.73% 3.91% 0.91% 1.01% 23.17% 27.32%
Moderate Sample (M=383)
Mean absolute deviation of weight changes across all assets 4.23% 3.26% 0.91% 1.06% 10.33% 15.49%
Active Share (AS) 6.35% 4.90% 1.36% 1.60% 15.50% 23.23%
AM 7.90% 5.81% 1.30% 1.63% 21.47% 26.52%
Conservative Sample (M=237)
Mean absolute deviation of weight changes across all assets 4.46% 3.44% 0.94% 1.07% 10.65% 12.99%
Active Share (AS) 6.70% 5.16% 1.41% 1.61% 15.97% 19.49%
AM 8.52% 6.73% 1.33% 1.53% 21.12% 23.24%
 Table 2, continued

Characteristics of Funds in Sample

Sample Sample Percentile

Sample Fund Characteristics (M=732) Mean Median 2.5% 5.0% 95.0% 97.5%

Panel C
Sample (M=732)
Mean absolute deviation of return differences for equities 2.76% 2.78% 1.82% 2.16% 3.44% 3.54%
Mean absolute deviation of return differences for bonds 2.66% 2.73% 1.76% 1.90% 3.39% 3.46%
Mean absolute deviation of return differences for cash 1.35% 1.36% 0.95% 1.07% 1.56% 1.60%
Mean absolute deviation of return differences across all assets 2.26% 2.27% 1.65% 1.78% 2.76% 2.86%
RD 3.08% 3.13% 2.13% 2.35% 3.78% 3.88%

Average of return differences for equities 0.15% 0.15% -0.50% -0.40% 0.73% 1.05%
Average of return differences for bonds 0.19% 0.19% -0.59% -0.28% 0.55% 0.62%
Average of return differences for cash -0.34% -0.34% -0.71% -0.65% -0.08% 0.00%

Panel D
Sample (M=732)
Number of asset allocation observations 63.7 56.0 8.0 11.0 139.0 152.5
Number of months measured 124.5 106.0 18.3 23.6 290.5 342.7
Months between observations 2.2 2.1 1.0 1.0 3.9 4.5
Fund size - dollars invested (in millions) $697.9 $115.8 $3.1 $7.4 $2,601.7 $5,766.7
 Table 3

Measures of Performance and Components of Funds in Sample

Table 3 reports on our measures of mutual fund performance and their corresponding performance components for
the 732 mutual funds in the sample. Panel A reports the distributions of the Active Returns (AR), Forecasting Success
(FS), Active Management (AM), and Return Dispersion (RD) for the funds in the sample as a whole and for three
Morningstar categories. Panel B shows the correlations between AM versus AS (Active Share measure of Cremers
and Petajisto (2009)), which also equals 3/2 times MADW. Note that AS and MADW, which is mean absolute
deviation of the weight changes, have a correlation of one. Panel B also reports the correlations between RD and
MADR, which is the mean absolute deviation of return differences.

Sample Sample Percentiles

Sample Fund Characteristics (M=732) Mean Median 2.5% 5.0% 95.0% 97.5%

Panel A
Sample (M=732)
Average Monthly Excess Return, AR = (rp - rb) 0.01% 0.00% -0.16% -0.10% 0.12% 0.21%
annualized AR 0.10% 0.04% -1.86% -1.15% 1.42% 2.56%
AR fund t-stat 0.14 0.17 -2.06 -1.63 1.80 2.02
FS 0.0111 0.0083 -0.1665 -0.1323 0.1546 0.2072
AM 7.92% 5.83% 1.19% 1.46% 21.53% 27.03%
RD 3.08% 3.13% 2.13% 2.35% 3.78% 3.88%

Aggressive Sample (M=112)

Average Monthly Excess Return, AR = (rp - rb) 0.00% 0.00% -0.13% -0.09% 0.08% 0.14%
annualized AR -0.02% 0.00% -1.58% -1.10% 1.01% 1.69%
AR fund t-stat 0.05 -0.02 -2.06 -1.51 1.71 1.86
FS 0.0017 -0.0010 -0.1645 -0.1438 0.1435 0.1752
AM 6.73% 3.91% 0.91% 1.01% 23.17% 27.32%
RD 3.19% 3.18% 2.52% 2.73% 3.72% 3.79%

Moderate Sample (M=383)

Average Monthly Excess Return, AR = (rp - rb) 0.01% 0.01% -0.15% -0.09% 0.11% 0.14%
annualized AR 0.11% 0.07% -1.77% -1.09% 1.34% 1.71%
AR fund t-stat 0.17 0.20 -2.04 -1.64 1.78 1.94
FS 0.0112 0.0100 -0.1607 -0.1232 0.1482 0.1934
AM 7.90% 5.81% 1.30% 1.63% 21.47% 26.52%
RD 3.01% 3.06% 2.08% 2.30% 3.78% 3.90%

Conservative Sample (M=237)

Average Monthly Excess Return, AR = (rp - rb) 0.01% 0.01% -0.19% -0.10% 0.21% 0.29%
annualized AR 0.16% 0.08% -2.23% -1.21% 2.51% 3.55%
AR fund t-stat 0.15 0.19 -2.03 -1.60 1.91 2.13
FS 0.0153 0.0100 -0.1732 -0.1245 0.1714 0.2180
AM 8.52% 6.73% 1.33% 1.53% 21.12% 23.24%
RD 3.12% 3.16% 2.16% 2.44% 3.78% 3.83%
 Table 3, continued

Measures of Performance and Components of Funds in Sample

Panel B
Sample (M=732) Aggressive (M=112)
Pearson Spearman Pearson Spearman
Correlations Between: Correlation Correlation Correlation Correlation
AM and AS (= 3/2 * MADW) 0.9073 *** 0.9190 *** 0.8839 *** 0.9496 ***

RD and MADR 0.9860 *** 0.9814 *** 0.9908 *** 0.9785 ***

Moderate (M=383) Conservative (M=237)

Pearson Spearman Pearson Spearman
Correlations Between: Correlation Correlation Correlation Correlation
AM and AS (= 3/2 * MADW) 0.9186 *** 0.9265 *** 0.9038 *** 0.8924 ***

RD and MADR 0.9855 *** 0.9830 *** 0.9859 *** 0.9771 ***
 Table 4

Selected Fund Comparisons

Table 4 contains comparisons of selected funds to illustrate the information contained in the Forecasting Success (FS),
Active Management, AM, and Return Dispersion, RD, components of performance as measured by Active Returns
(AR). Panel A shows two comparisons to illustrate the effect of FS on AR. Panel B shows two comparisons to illustrate
the effect of AM on AR. Panel C shows two comparisons to illustrate the effect of RD on AR.
Annualized
Sample Comparison Funds AR AR FS AM RD

Panel A - FS

Differing AR due to FS differences (when FS > 0)

Fund A1 (Higher AR Fund due to a higher FS) 0.35% 4.23% 0.2210 15.52% 3.36%
Fund A2 (Lower AR Fund due to a lower FS) 0.07% 0.90% 0.0490 15.35% 3.32%
Difference 0.27% 3.33% 0.1720 0.17% 0.04%

Differing AR due to FS differences (when FS< 0)

Fund A3 (More negative AR Fund due to a more negative FS) -0.52% -6.07% -0.1577 31.60% 3.48%
Fund A4 (Less negative AR Fund due to a less negative FS) -0.16% -1.90% -0.0563 31.80% 2.97%
Difference -0.36% -4.18% -0.1014 -0.21% 0.52%

Panel B - AM

Differing AR due to AM differences (when FS > 0)

Fund B1 (Higher AR due to a higher AM multiplying a positive FS) 0.29% 3.55% 0.1272 23.47% 3.25%
Fund B2 (Lower AR due to a lower AM multiplying a posiitive FS) 0.03% 0.40% 0.1275 2.64% 3.28%
Difference 0.26% 3.15% -0.0003 20.83% -0.03%

Differing AR due to AM differences (when FS < 0)

Fund B3 (More negative AR due to a higher AM multiplying a negative FS) -0.52% -6.07% -0.1577 31.60% 3.48%
Fund B4 (Less negative AR due to a lower AM multiplying a negative FS) -0.07% -0.78% -0.1598 3.91% 3.48%
Difference -0.46% -5.29% 0.0021 27.68% 0.01%

Panel C - RD

Differing AR due to RD differences (when FS > 0)

Fund C1 (Higher AR due to a higher RD multiplying a positive FS) 0.05% 0.64% 0.1094 4.24% 3.82%
Fund C2 (Lower AR due to a lower RD multiplying a positive FS) 0.04% 0.53% 0.1096 4.24% 3.18%
Difference 0.01% 0.11% -0.0002 0.01% 0.65%

Differing AR due to RD differences (when FS < 0)

Fund C3 (More negative AR due to a higher RD multiplying a negative FS) -0.09% -1.09% -0.0901 11.28% 2.99%
Fund C4 (Less negative AR due to a lower RD multiplying a negative FS) -0.07% -0.82% -0.0948 11.38% 2.12%
Difference -0.02% -0.27% 0.0047 -0.11% 0.87%
 Table 5

Subsample Regressions splitting each funds sample period

Table 5 shows regression coefficients and t-statistics on the 678 funds with at least 26 observations. These regressions
examine persistence in Active Returns (AR), Forecasting Success (FS), and Active Management (AM), which are the
elements of performance under a fund’s control. The subperiods represent the first and second halves of the
observations for each fund. AR1, FS1, AM1, and RD1 are the Active Returns, Foresight, Active Management, and
Return Dispersion in the first half subperiod. *, **, *** indicate significance at the 10%, 5%, and 1% level.

Full Sample (M=678)

Variable AR2 AR2 FS2 AWD2 AWD2

Intercept -0.0001 0.0029 0.0435 *** 0.0355 ***

(-0.60) (0.10) (4.70) (3.64)

FS1 0.0005 0.1334 *** -0.0327 ** 0.0513

(1.28) (3.20) (-2.46) (1.64)

AM1 0.0007 0.0630 0.2006 *** 0.2071 ***

(0.96) (0.77) (7.70) (7.96)

RD1 0.0047 -0.1775 -0.3883 * -0.5016 *

(0.58) (-0.19) (-1.92) (-1.70)

D1 0.0102 *
(1.69)

D1 * FS1 -0.1045 ***

(-2.60)

AR1 0.0424
(1.27)

adjusted R-squared 0.0000 0.0001 0.0115 0.0902 0.0993