THREE ESSAYS ON MUTUAL FUNDS

Saurin Patel

Desautels Faculty of Management,

McGill University,

Montreal, Quebec, Canada

June 2013

A thesis submitted to McGill University in partial fulfillment of the

requirements for the degree of Doctor of Philosophy

Copyright © 2013 by Saurin Patel. All rights reserved.

2

DEDICATION

To my beloved family.

3

ACKNOWLEDGEMENTS

I have many individuals to thank for their help during the course of my doctoral

studies. First and foremost, I want to thank my co-supervisors, Professors Sergei

Sarkissian and Susan Christoffersen, for their guidance, support, and mentoring over the

years. I appreciate all their contributions of time, ideas, and funding to make my doctoral

experience productive and stimulating. Their comments and feedback on the chapters of

this thesis were extremely helpful and showed me the path to become a better researcher.

Without both of them, I would not be where I am academically.

I thank my dissertation committee members, George Aragon, Laurent Barras,

Martin Boyer, Dr. John Burgess, Abhirup Chakrabarti, and Jean-Claude Cosset for

agreeing to be a part of my committee and allocating their time to providing valuable

feedback.

I am very grateful to the administrative staff in the Ph.D. office for their constant

support. In particular, I want to thank Stella Scalia, Susan Lovasik, and Karen Robertson

for their thoughtfulness and eagerness to help. A special thanks to Pierre Cambron for his

invaluable help in keeping my computers running smoothly.

I would also like to express my gratitude to Professor Saibal Ray, Ph.D. Program

director, for his constant encouragement and support throughout my doctoral studies.

I gratefully acknowledge the financial support from McGill University, Fond

Québécois de Recherche sur la Société et Culture (FQRSC), Walter John Stenason Ph.D.

Fellowship and National Bank Financial Group Ph.D. Fellowship. Without their

generous financial support, it would have been impossible for me to complete my

doctoral studies.

I want to thank my fellow doctoral students, many of whom have become very

close friends for life. Especially Burcin Col, Paul Intrevado, Rajasekhar Kakumani,

Mehdi Karoui, Xuhui (Nick) Pan, and Russell Seidle, who have taught me a great deal

about finance and more importantly about academic life. I have thoroughly enjoyed their

camaraderie and will never forget the quality time I have spent with them.

4

Last but not least, I want to thank my family for all their unconditional love and

constant encouragement. Without them, none of this would be possible. Especially my

parents, Kusum and Pramod Patel, who have taught me the true meaning of life and have

always encouraged me to take on difficult challenges in life including pursuing a doctoral

degree. My sisters, Urvashi and Jignya, who have always believed in me and showed me

the right path in life. Finally, my loving wife, Pinal, whose unceasing support during the

final stages of my doctoral studies helped me complete this dissertation.

5

ABSTRACT

This dissertation consists of three essays. The first essay examines whether investors' optimism about the future economic growth affects their future investment decisions. Drawing from the insights of the theoretical literature on investment behavior, we argue that investors base their future investment decisions not only on asset-specific information and existing macroeconomic environment, but also on their beliefs about future economic growth. Consistent with this premise, we find that investors' investment decisions, measured by mutual fund flows, are positively influenced by their economic optimism, even after controlling for various fund characteristics and macroeconomic conditions. Moreover, this effect is more pronounced for funds with greater fund-specific information uncertainty, i.e. funds that are small, belong to smaller fund families, or have highly volatile past performance. Our results suggest that investors not only consider forward-looking economic optimism in their investment decisions, but also put greater weight on it when fund-specific information seems uninformative and less valuable.

The second essay examines the impact of group decision making on fund performance, their risk-taking behavior and other fund characteristics using a large U.S. equity mutual fund database. The literature has conflicting reports regarding the impact of team based managerial structure on fund performance. We first observe that in mutual fund studies this results from large discrepancies in reported managerial structures between CRSP and Morningstar databases reaching on average 20% per year. Then we show that with more superior Morningstar data team-managed funds exhibit higher risk-adjusted returns than single-managed funds. The performance spread is present across all fund categories, except aggressive funds, and is robust to the inclusion of fund- and manager-level controls. Across various managerial structures, the largest team-induced gains are reached among funds managed by three individuals. Furthermore, teams significantly improve fund performance when funds are located in financial centers, reflecting larger networking potential and/or better skills of people who reside in larger cities. This improvement is achieved in teams more homogeneous in age and education. In spite of higher returns however, team-managed funds are not riskier than single-managed funds in terms of market exposure or idiosyncratic volatility. Finally, team-managed funds trade less aggressively and are able to generate extra inflows for their funds. Thus, collective decision making is beneficial, but its scale depends on team size and diversity as well as its geographic location.

The third essay examines the relation between managerial structure and the likelihood of deception. Using U.S. equity mutual fund data, we find that team-managed funds deceive significantly less than single-managed funds. In particular, we show that two trading activities – portfolio pumping and window dressing, – which are considered illegal or quasi-illegal, are more profound or exists at all only among single-managed funds. We also document a negative relation between the extent of those two activities and team size. Subsequent tests indicate that these results are not driven by various fund characteristics that differ between single- and team-managed funds, such as fund returns, size, and turnover. In addition, we observe that portfolio pumping is present most strongly among the worst performing single-managed funds, while window dressing

6

occurs primarily again among single-managed funds but in the middle performance group for which it has the largest potential benefits. Overall, our findings support the notion that team is a desirable form of organization as it helps weaken incentives to deceive.

7

RÉSUMÉ

Cette thèse se compose de trois essais. Le premier essai examine si l’optimisme des investisseurs concernant la croissance économique future affecte leurs décisions d’investissement futures. En s’appuyant sur les connaissances de la littérature théorique sur le comportement des investisseurs, nous pensons que les investisseurs fondent leurs décisions d’investissement futures non seulement sur l’information spécifique aux actifs et l’environnement macro-économique, mais aussi sur leurs croyances au sujet de la croissance économique future. En accord avec cette hypothèse, nous trouvons que les décisions d’investissement, mesurée par les flux de fonds communs de placement, sont positivement influencées par l’optimisme économique des investisseurs, même après avoir contrôlé pour les caractéristiques diverses des fonds et les conditions macroéconomiques. De plus, cet effet est plus prononcé pour les fonds avec une incertitude plus grande de l’information propre au fonds, c’est-à-dire les fonds qui sont petites, qui appartiennent à des familles de fonds plus petits, ou qui ont des performances passées très volatiles. Nos résultats suggèrent non seulement que les investisseurs tiennent compte des énoncés économiques prospectifs de caractère optimiste dans leurs décisions d’investissement, mais qu’ils mettent aussi plus de poids sur ces énoncés quand l’information spécifique au fonds semble peu informative et moins valable.

Le deuxième essai examine l’impact de la prise de décision collective sur le rendement du fonds, leur comportement en termes de prise de risque et autres caractéristiques de fonds, en utilisant un grand base de données traitant des fonds communs de placement américains. La littérature démontre des rapports contradictoires concernant l’impact de la structure de gestion d’équipe sur le rendement des fonds. Nous observons tout d’abord que dans les études de fonds communs de placement ce résultat découle des grandes différences dans les structures de gestion rapportés entre les bases de données CRSP et Morningstar, qui atteignant en moyenne 20 pourcent par an. Ensuite, nous montrons qu’avec les données supérieures de Morningstar les fonds gérés par des équipes présentent des rendements ajustés au risque plus élevés que celles des fonds gérés par les individus. L’écart de performance est présent à travers toutes les catégories de fonds, à l’exception des fonds agressifs, et est robuste à l’inclusion des contrôles au niveau du fonds et du gestionnaire. À travers diverses structures de gestion, les plus grandes gains induites par les équipes sont atteints par les fonds gérés par trois individus. Par ailleurs, les équipes améliorent considérablement le rendement du fonds lorsque les fonds sont situés dans des centres financiers, reflétant le plus grand potentiel de réseautage et / ou les meilleures compétences des personnes qui habitent dans les grandes villes. Cette amélioration est obtenue au sein d’équipes plus homogènes en termes d’âge et de l’éducation. En dépit des rendements plus élevés, toutefois, les fonds gérés par des équipes ne sont pas plus risqués que les fonds gérés par un seul individu en termes d’exposition au marché ou à la volatilité idiosyncrasique. Enfin, les fonds gérés par des équipes s’échangent de façon moins agressive et sont capables de générer des entrées supplémentaires pour leurs fonds. Ainsi, la prise de décision collective est bénéfique, mais son ampleur dépend de la taille de l’équipe et de la diversité ainsi que de sa situation géographique.

8

Le troisième essai examine la relation entre la structure de gestion et le risque de tromperie. En utilisant les données sur les actifs de fonds communs de placement américains, nous constatons que les fonds gérés par des équipes trompent beaucoup moins que ceux gérés par les individus. En particulier, nous montrons que deux activités – ‘portfolio pumping’ et ‘window dressing’ – qui sont considérées comme illégales ou quasi-clandestines, sont plus profondes ou existent seulement parmi les fonds gérés par les individus. Nous avons également documenté une relation négative entre l’ampleur de ces deux activités et la taille de l’équipe. Des tests ultérieurs indiquent que ces résultats ne sont pas motivés par diverses caractéristiques des fonds qui diffèrent entre les fonds gérés par individus et équipes, tels que le rendement des fonds, la taille, et le chiffre d’affaires. En outre, nous observons que le ‘portfolio pumping’ est présent le plus fortement parmi les fonds moins performants gérés par les individus, tandis que le ‘window dressing’ se produit principalement chez les fonds gérés par les individus, mais dans le groupe de performance moyenne pour lequel il présente les plus grands avantages possibles. Dans l'ensemble, nos résultats soutiennent l’idée que l’équipe est une forme d’organisation souhaitable car elle contribue à affaiblir les incitations à tromper.

9

Table of Contents

Dedication ....................................................................................................................................... 2

Acknowledgements ........................................................................................................................ 3

Abstract .......................................................................................................................................... 5

Résumé ........................................................................................................................................... 7

Contribution of Authors ............................................................................................................. 11

Chapter 1 ...................................................................................................................................... 12 Introduction

Chapter 2 ...................................................................................................................................... 18 Economic Optimism, Information Uncertainty and Future Investment Decisions: Evidence of the Mutual Fund Industry

2.1 Introduction ...................................................................................................................................................19

2.2 Hypotheses Development ..............................................................................................................................23

2.3 Data ...............................................................................................................................................................29

2.3.1 Investor Expectations ...........................................................................................................................29

2.3.2 Mutual Fund Sample ............................................................................................................................33

2.3.4 Macroeconomic Control Variables ......................................................................................................37

2.4 Empirical Analysis ........................................................................................................................................38

2.4.1 Influence of Investor Expectations on Future Fund Flows ..................................................................38

2.4.2 Role of Information Uncertainty in Fund Purchase Decisions .............................................................42

2.5 Summary and Conclusion ..............................................................................................................................49

Appendix .................................................................................................................................................................51

Figure ......................................................................................................................................................................54

Chapter 3 ...................................................................................................................................... 65 To Group or Not to Group? Evidence from Mutual Funds

3.1 Introduction ...................................................................................................................................................66

3.2 Motivation and Hypotheses Development .....................................................................................................71

3.3 Data ...............................................................................................................................................................74

3.3.1 Main Data Source ................................................................................................................................74

3.3.2 Fund Characteristics .............................................................................................................................75

3.3.3 Fund Manager Characteristics .............................................................................................................76

3.3.4 Fund Performance Measures ................................................................................................................78

3.3.5 Summary Statistics...............................................................................................................................80

3.4 Management Structure: CRSP versus Morningstar .......................................................................................82

3.4.1 Fund Management Structure Differences ............................................................................................82

3.4.2 Fund Performance Differences ............................................................................................................85

3.4.3 Additional Misspecification Issues in Management Structure .............................................................88

3.5 Team Management and Fund Performance: Empirical Tests ........................................................................89

3.5.1 The Average Effect of Team Management ..........................................................................................89

3.5.2 The Effect of an Additional Team Member .........................................................................................92

3.5.3 Team Management and Geographic Location .....................................................................................94

3.5.4 The Role of Team Diversity ................................................................................................................96

3.6 Team Management, Risk Taking, and Fund Characteristics .........................................................................99

3.7 Conclusions .................................................................................................................................................102

Table .....................................................................................................................................................................104

10

Chapter 4 ................................................................................................................................... 119 Deception and Managerial Structure: A Joint Study of Portfolio Pumping and Window Dressing Practices

4.1 Introduction ................................................................................................................................................. 120

4.2 Motivation and Related Literature............................................................................................................... 125

4.3 Portfolio Pumping and Managerial Structure .............................................................................................. 129

4.3.1 The Detail of Portfolio Pumping Phenomenon and its Estimation Methodology .............................. 129

4.3.2 Test Results ........................................................................................................................................ 133

4.4 Window Dressing and Managerial Structure ............................................................................................... 138

4.4.1 The Details of Window Dressing Phenomenon and its Estimation Methodology ............................. 138

4.4.2 Test Results ........................................................................................................................................ 142

4.4.3 The Dot-com Bubble: A Special Case of Window Dressing ............................................................. 148

4.5 Conclusions ................................................................................................................................................. 149

Appendix............................................................................................................................................................... 151

Table ..................................................................................................................................................................... 154

Chapter 5 ................................................................................................................................... 168 Conclusions

Bibliography .............................................................................................................................. 173

11

CONTRIBUTION OF AUTHORS

The second chapter of this thesis titled, "Economic Optimism, Information

Uncertainty and Future Investment Decisions: Evidence from the Mutual Fund Industry",

is my work and I am solely responsible for this chapter.

The third and fourth chapters of this thesis titled, "To Group or Not to Group?

Evidence from Mutual Funds", and, "Deception and Managerial Structure: A Joint Study

of Portfolio Pumping and Window Dressing Practices", respectively, are in joint

collaboration with Sergei Sarkissian. Currently, Sergei is an Associate Professor of

Finance at Desautels Faculty of Management, McGill University. We have contributed

equally to both these chapters.

12

Chapter 1

Introduction

13

Over the past several decades, the U.S. mutual fund industry has experienced

tremendous growth. Right from $135 billion in total assets under management at the end

of 1980 to $13 trillion in assets at the end of 2011. One of the main drivers behind this

remarkable growth is the increased use of mutual funds as a preferred investment vehicle

by individual households in United States. According to 2012 Investment Company

Institute (ICI) Fact book, an estimated 52.3 million households, or 44 percent of all U.S.

households, owned mutual funds in 2011 compared to only 4.6 percent in 1980. The

surge in the U.S. mutual fund industry size along with the increased participation of

individual households provides a unique opportunity to study investment behavior of

individual households. In particular, what type of information is relevant to households

and how they make their mutual fund investment decisions?

Understanding these questions is at the heart of chapter two. Economic theory

argues investors base their investment decisions broadly on three types of information.

First, the asset-specific information such as past dividends, valuation ratios and other

characteristics that signals the intrinsic value of an asset. Second, the information related

to the current economic conditions that reflects the existing investment opportunities.

And third, investors' expectations about the future economic environment which might

affect the future returns of an asset. While much of previous literature, especially in

mutual funds, focuses on fund- and economy-specific information to explain mutual fund

flows, a proxy for investors' investment decisions. Little attention is given to whether

expectations affect investment decisions and if so, how expectations enter the investment

decision-making process. Addressing these questions is the main objective of chapter

two. Building on the theoretical insight, we empirically test if expectations about future

economic growth can influence investors' investment decisions, in addition to other well

known information variables. We start by constructing a new and robust measure of

investor expectations (INVEXP) which accurately captures individual investors'

14

expectations about what will happen to the economy as a whole in the future. Unlike

existing proxies, our measure offers three distinct advantages: i) it is a direct and cleaner

measure, ii) it is highly correlated to well known state variables, and iii) it has desirable

statistical properties such as relative stability, stationarity and persistence. We then

investigate whether INVEXP affects individual mutual fund flows, a proxy of investor's

investment decisions. We find that INVEXP positively influences mutual fund flows.

This result shows that expectations are important to investors and investors increase their

allocation to riskier equity fund when they are optimistic about the future. Then, we

focus on how expectations enter the investment decision-making process. To test this we

follow the economic theory which suggests investors trade-off different pieces of

information in making their investment decisions based on the quality or precision of

information.1 We find that mutual fund flows are more sensitive to expectations for funds

that have noisy and poor quality of fund-specific information, such as funds with highly

volatile past performance, funds that belong to smaller fund families and smaller sized

funds. Overall, our results strongly indicate the positive influence of investors' economic

expectations on their investment decisions and show that investors put greater weight on

their economic expectations in making investment decisions whenever the fund-specific

information is noisy and less informative.

Another important yet understudied trend in the U.S. mutual fund industry is the

rise of team-managed funds. In 1992, approximately 30 percent all domestic diversified

equity mutual funds were managed by teams of portfolio managers whereas 70 percent

managed by single portfolio managers. Compare that to 2010, when approximately 70

percent of all equity funds were managed by teams and only 30 percent by single

managers. This sudden rise in team-based managerial structure raises an important

question: Why are mutual fund families moving towards team-based management?

1 See e.g. Grossman and Stiglitz (1980).

15

Chapter three investigates the rise of team-based management structures from the

perspective of fund performance. Theoretical studies, such as Sharpe (1981) and Barry

and Stark (1984), argue that teams are beneficial in the portfolio management industry

because they improve fund performance by achieving higher degree of specialization,

diversification of opinions and reducing portfolio risk. But several empirical studies find

that team-managed funds perform poorly compared to single-manager funds.2 This

evidence is puzzling because it is in direct conflict with theory and industry trend. In

chapter three, we examine this puzzle in the light of new data. Using a relatively new

data from MorningStar, we find that there exists a positive relation between teams and

fund performance, consistent with the theory and industry trend. Further, we show that

the negative relation documented by previous empirical studies is because of large

misspecification in CRSP managerial data. We find that CRSP database inaccurately

reports the number of portfolio managers responsible for day to day activities of a fund.

This misspecification ranges between as low as 10 percent to as high as 26 percent of

entire sample per year. The second part of this chapter analyzes whether number of team-

members, location of teams and diversity among team-members affects the team-fund

performance relation. No empirical mutual fund study to-date, to the best of our

knowledge, analyzes these conditional effects. We show that there exists a non-linear

relation between team size and fund performance. In particular, we find that three-

member teams tend to generate highest fund performance relative to single-manager

funds. We also show that location of teams plays a very important role in teams'

performance. We find the benefits of team management are strongly present among

funds located in financial centers but not among those located in non-financial centers.

We also show that diversity among team-members, especially in age and educational

backgrounds, can hurt team performance. In other words, funds with more homogeneous

2 See e.g. Chen, Hong, Huang and Kubik (2004), Kempf and Ruenzi (2005) and Han, Noe and Rebello (2008).

16

team members in terms of age and educational background outperform those with more

heterogeneous managers. Overall, our results point to the benefits of team-based

managerial structures in terms of fund performance which might help explain the rise of

teams within the mutual fund industry.

Chapter four investigates the relation between team-based managerial structure

and the likelihood of engaging in unethical or illegal conduct. Discouraging people from

engaging in unethical or illegal behavior remains a big challenge for social sciences

especially economics. The economic literature points to two reasons which encourage

agents to engage in unethical behavior: i) over-incentivization within contracts (see e.g.

Jacobs and Levitt (2003)) and ii) imperfect monitoring (see e.g. Holmstrom (1979)). But

at the same time, several theoretical studies argue that "team production" can address

both these reasons by reducing the "over-incentivization" problem and increasing

monitoring through peer monitoring.3 We test this theoretical assertion empirically in this

chapter. In particular, we test if team-based managerial structures deter agents from

engaging in deceptive and unethical behavior. We achieve this goal by examining the

extent of portfolio pumping and window dressing -- the two fund trading practices that

are considered illegal and quasi-illegal respectively -- among single- and team-managed

funds in the U.S. mutual fund industry. Consistent with prior theoretical studies, we show

that single-managed funds involve in these deceptive trading practices significantly more

than team-managed funds. In particular, we find that team-managed funds involve

significantly less in portfolio pumping and do not involve in window dressing at all.

Moreover, we also document a negative relation between the extent of these two

dishonest trading activities and team size. Overall, our results show that team-based

managerial structure can significantly inhibits managers' drive to engage in dishonest

3 See Arnott & Stiglitz (1991), Kandel & Lazear (1992) and Acemoglu, Kremer and Main (2008).

17

trading practices which can be beneficial to both individual investors and industry as a

whole.

18

Chapter 2

Economic Optimism, Information

Uncertainty and Future Investment Decisions:

Evidence from the Mutual Fund Industry

19

"The task of security analysis consists in rearranging the data descriptive of the state of

the economy, of the states and trends of the industries of specific firms, and of firms

themselves, [...] in the light of the bases on which investors individually may judge

prospects. Whatever tools of analysis an investor uses, [...] the only appropriate basis on

which to found expectations and judgments is the composite set of environmental and

financial facts available at the time of choice."

Smith (1971)

2.1 Introduction

Making investment decisions is a complex task. It involves gathering and

analyzing a large amount of information in order to make an informed choice. But little is

known about the precise nature of information that investors gather and how they analyze

this information in their minds? In this paper, we broadly examine the different types of

information investors gather and how this information enters their investment decision.

Investors care about information that helps them assess the future value of an

investment. Intuitively, this information can be broadly divided into three types. First, the

asset-specific information such as past dividends, valuation ratios and other

characteristics that signals the intrinsic value of an asset. Second, the information related

to the current economic conditions that reflects the existing investment opportunities.

And third, the anticipation or optimism about future economic environment which might

affect the future returns of an asset. Financial theory acknowledges this intuition and

describes investors' information set as a function of observable asset-specific

information, current state of economy, and expected economic and political environment

(e.g., Smith (1971), Veronesi (2000)). Particularly, emphasizing the importance of

investors' economic expectations or beliefs in investment decisions (see e.g., Shackle

(1942), Angell (1960), Smith (1971) and Veronesi (2000)). However, the empirical

literature focuses only on asset-specific and current economic information to explain

20

investor decisions and pays little attention to the role of investors' economic optimism.4

While few studies recognize that flows into mutual funds capture investors' optimism and

sentiment, the literature does not separate the part of funds flows arising from optimism

compared to asset-specific and current macroeconomic information.5

This leads us to our first research question. Does optimism about future economic

growth play an important role in investors' investment decision-making in addition to

asset-specific and current economic information? The next question that arises is - how

do investors combine these three types of information to arrive at a final decision?

Financial theory argues that investors combine different pieces of information based on

the quality of the information (see e.g., Grossman and Stiglitz (1980), Veronesi (2000)

and Epstein and Schneider (2008)). That is, investors put less weight on the information

that has relatively lower quality, and more weight on relatively higher quality

information. For instance, in situations when investors are uncertain about investing in an

asset based on its noisy asset-specific information, they tend to rely more on other pieces

of information in their information set to make an investment decision. Similarly, when

investors are absolutely certain about the value of asset based on high-quality asset-

specific information, they pay less attention to other information in their information set.

Following this insight, we test the following research question: Do investors tradeoff

between different types of information based on information quality when making

investment decision?

4 See e.g., Ippolito (1992), Hendricks, Patel and Zeckhauser (1993), Chevalier and Ellison (1997), Sirri and Tufano (1998), Grinblatt and Keloharju (2000) and Barber, Odean and Zheng (2005) among others show the influence of asset-specific information of investor decisions; Malmendier and Nagel (2011) show influence of stock market's past performance on individual investors decisions; Warther (1995), Aber and Santini (1998), Cao, Chang and Wang (2008) and Chalmers, Kaul and Phillips (2011) show the influence of current economic conditions on aggregate investment decisions.

5 See Ben-Rephael, Kandel and Wohl (2011) for references.

21

To test both these questions empirically, we use the mutual fund industry data in

the United States along with a novel measure of investors' economic optimism. Mutual

fund industry is one of the best places to study investment behavior because one can with

relative ease and clarity infer the investment decisions of investors through flows into

and out of individual mutual funds (see Ippolito (1992), Sirri and Tufano (1998)). In

addition, the data on mutual fund flows goes back enough in time that researchers are

able to study investor behavior over a significantly longer time periods. And it is

relatively straightforward to define and control asset-specific information in a mutual

fund setting using different fund characteristics such as fund performance and fund fees.

To capture investors' optimism about future economic growth, we use the survey

data from the University of Michigan Survey Research Center (henceforth UMSRC).

Every month UMSRC surveys individual households on their expectations about future

economic conditions in the U.S. including business conditions (including future stock

market growth, GDP growth, etc.), inflation, unemployment, interest rates as well as

their personal financial situation. Using the survey responses that focus solely on

economic expectations, we construct a new expectations index (INVEXP). Using

quarterly mutual fund data from 1970 to 2008, we find that investors' investment

decisions, proxied by fund flows, respond positively to optimism about future economic

conditions even after controlling for fund-specific and current economic information.

That is, one standard deviation increase in investors' optimism increases fund flows into

an average equity fund by 4% (or $12 million) per year in addition to fund-specific and

current economic conditions.6 This result clearly shows the importance of investors'

economic optimism in their investment decisions and validates the theoretical insight

provided in the literature.

6 Compared to $118 million of new money inflows per year in an average sized fund in our sample.

22

To test whether investors tradeoff different pieces of information in making

investment decision, we find that when fund-specific information is noisy and of poor-

quality, investors rely more on their economic optimism in making their investment

decisions. That is, flows into funds with noisy fund-specific information are more

sensitive to economic optimism than funds with relative more precise fund-specific

information. Using different fund characteristics as proxies for high fund-specific

information uncertainty, we find that flows into funds with highly volatile past

performance, funds that belong to smaller fund families and smaller sized funds receive

additional inflows ($9 - $19 million per year) when investors are optimistic about future

economic conditions. These results clearly show that investors not only consider their

economic optimism in making investment decisions, but also put greater weight on it

whenever the asset-specific information is noisy or less precise.

Our paper makes three contributions to the literature. First, it offers novel

approach to understanding investors' investment decisions based on three different types

of information: asset-specific, current economic conditions, and expectations about

future economic conditions. Unlike previous studies, we examine the role of investors'

economic optimism in investors' investment decisions in the presence of asset-specific

information and current economic conditions. We find robust evidence of positive and

significant relation between investors' economic optimism and their future investment

decisions. That is, investors not only consider forward-looking economic information in

addition to asset-specific and existing economic information before making their

investment decisions, but also increase (decrease) their wealth allocation into risky assets

when they are optimistic (pessimistic) about the future.

Second, we offer new empirical evidence on how investors combine different

types of information when making investment decisions. Consistent with theoretical

literature, we find that investors tradeoff different types of information based on the

23

quality of information. In particular, we find that investors put greater weight on their

economic optimism in making investment decisions when the fund-specific information

is noisy and less informative.

Finally, we develop a new and robust measure of investor expectations

(INVEXP) that captures investors' optimism/pessimism about future economic

conditions. Unlike the existing expectations proxies, this proxy offers three distinct

advantages: (i) it is a direct and cleaner measure of investor expectations, (ii) it is highly

correlated with known state variables which predict future economic conditions, and (iii)

it has desirable statistical properties such as relative stability, stationarity and lines up

accurately with known periods of investor optimism and pessimism.

The rest of this paper is organized as follows. Section II discusses the related

literature and hypotheses development. Section III describes, in detail, the construct and

validation of the investor expectations proxy along with the mutual fund data. Section IV

covers the empirical test design that investigates the link between investor expectations

and mutual fund flows and presents empirical evidence as to why this link may exist.

Section V summarizes and concludes.

2.2 Hypotheses Development

Several theoretical studies in finance recognize the importance of expectations or

optimism in investors' investment decision-making process. Studies as early as Shackle

(1942) point out that economic theory places too much emphasis on investors'

preferences and too little on their feelings of hope, doubt or fear. Shackle argues that

investors base their investment decisions on the notion of "potential surprise" which is

the degree of surprise beyond investors' initial expectations. Angell (1960), in line with

24

Shackle (1942), argues that may prevail during the life of the investment in their

investment decisions. In doing so, investors judge the "prospect" for success of a

particular investment security under these various alternatives. Smith (1971) argues that

in order to make informed investment choice, investors not only need some knowledge

about the different characteristics of investment security, but also need to know the state

and trend of a nation's economy, as well as the conditions in the money and securities

markets. More recent studies such as Veronesi (2000) and Kacperczyk, Van

Nieuwerburgh and Veldkamp (2011) formalize the role of expectations in investment

decisions using standard general equilibrium framework and model investors' posterior

distribution as a function of investors' expectations about future economic growth and

asset-specific information.

Despite such strong theoretical arguments in favor of investor optimism, the

empirical literature overlooks investor optimism and focuses only on asset-specific and

existing macroeconomic information in explaining their investment decisions. For

instance, a large body of empirical literature in mutual funds shows that investors'

decisions to buy and sell mutual funds is influenced by fund-specific information such

as fund's past performance (see e.g., Ippolito (1992), Hendricks, Patel and Zeckhauser

(1993), Sirri and Tufano (1998)); fund fees (see Sirri and Tufano (1998) and Barber,

Odean and Zhang (2005)); and other non-performance characteristics such as fund

advertisements expenses (see Capon, Fitzsimons and Prince (1996); Jain and Wu

(2000)). Research from non-mutual fund setting also shows that individual household's

decisions to buy and sell individual stocks is influence by historical stock returns

(Grinblatt and Keloharju (2000)). Similarly, another strand of empirical literature shows

that investors' investment decisions, in aggregate, are influenced by existing financial and

economic conditions such as stock market returns (Warther (1995)); stock market

volatility (Cao, Chang and Wang (2008)); and changes in aggregate economic conditions

25

such as short-term interest rates, personal disposable income, default spreads and term

spreads (Aber and Santini (1998), Chalmers, Kaul and Phillips (2011)).

The absence of any empirical investigation into the importance of investors'

economic optimism in their investment behavior motivates this paper. The first research

question that we ask is: Does investors' optimism about future economic growth affect

their investment decisions? Intuitively, investors, at least in aggregate, alter the riskiness

of their portfolios in response to expected changes in economic conditions. In other

words, when investors are optimistic about the future, they are willing to undertake more

risk by investing in relatively riskier equity funds. Thus, we should expect a positive

relation between investor optimism and subsequent mutual fund flows. This intuition

leads us to our first hypothesis:

Hypothesis 1: Investor expectations about future economic conditions are

positively associated with mutual fund flows in subsequent periods.

The second question that we raise is: how do investors combine the different types of

information in their minds to arrive at an actual investment decision? Do investors

simply aggregate all pieces of information or weigh different information based on some

criteria? Intuition suggests that given investors' limited capacity to process a large variety

of information simultaneously, weighing different types of information based on its

quality, reliability or likelihood of occurrence seems more likely choice. Economic

theory seems to support this intuition. For example, Grossman and Stiglitz (1980) argue

that investors' demand for a risky asset varies with the quality of information that

investors have in their information set. In particular, the higher the quality or precision of

information, the greater the weight that information in investors' information set and

hence the greater the demand for that asset. Veronesi (2000) rationalizes investors'

information set as a precision-weighted function of past dividend realizations and

26

expectations about future economic growth and argues that information quality affects

equity premium. More recently, Epstein and Schneider (2008) use Bayesian setting to

study the effect of uncertain information quality on asset prices and model investors'

posterior distribution as a function mean dividend payoff, future aggregate economic

conditions and idiosyncratic shock to the asset. All in all, these studies point to one thing:

investors tradeoff high quality information for low quality information in making

investment decisions. In particular, investors put less weight on the information that has

relatively lower quality, and more weight on relatively higher quality information. In our

setting, this implies that when investors are uncertain about the fund quality based on

noisy fund-specific information, they tend to rely more on other pieces of information in

their information set. That is, when there is high information uncertainty related to fund-

specific information, economic optimism plays a greater role in investors' investment

decisions by making investors less risk averse. A direct way to test this argument is to

see if the flows into funds with high information uncertainty are more sensitive to

investors' optimism than flows into funds with low information uncertainty. This

motivates our second hypothesis:

Hypothesis 2: The sensitivity of mutual fund flows to investor expectations is

greater for funds with higher fund-specific information uncertainty.

Following Hirshleifer (2001) and Zhang (2006), we formally define information

uncertainty as the ambiguity in assessing the quality of a fund based on noisy fund-

specific signals. The imprecision or noisiness in fund-specific signals stems from two

sources: volatility of a fund's underlying fundamental value and poor quality fund

information.7 We use different fund characteristics as proxies for funds' underlying 7 In standard theoretical literature, an observed signal (s) is viewed as a fund's fundamental value (f), such as fund manager's ability to generate returns in excess of a benchmark measured by risk-adjusted fund performance, plus some noise, that is, si = fi + ei. The information uncertainty is measured by the variance or noisiness in the observed signal, var(si) = var(fi) + var(ei), where var(fi) reflects volatility of fund i's underlying fundamentals and var(ei) reflects quality of fund-specific information.

27

fundamentals as well as quality of information and develop two testable empirical

predictions.

Our first empirical prediction deals with volatility in a fund's fundamental value

measured by volatility in managerial skill. Mutual fund investors value a fund based on

the fund manager's ability to generate returns in excess of a benchmark. Thus, the most

important determinant of a fund's fundamental value is managerial skill. We define

managerial skill as the ability of the fund manager to generate fund returns in excess of

passive as well as non-passive benchmarks (Pastor and Stambaugh (2002)). Empirically,

we measure managerial skill by monthly risk-adjusted fund performance. Any

uncertainty in ascertaining managerial skill affects investors' perception about the future

value of the fund. Using the standard deviation of monthly risk-adjusted fund

performance over the past year as a proxy for managerial skill uncertainty (Pastor and

Stambaugh (2002)), we argue that investors perceive funds with high fund performance

volatility as ones with greater managerial skill uncertainty.8 Thus, the uncertainty mainly

arises from the volatility in fund's underlying fundamental value. At the same time,

economic optimism about the future economy lessens investor's risk aversion towards

these perceived risky funds. And as a result, they underweight fund-specific information

for funds with greater managerial skill uncertainty and overweight their economic

expectations when making future fund purchase decisions in these funds. Formally put,

our first empirical prediction is:

Prediction 1: The sensitivity of fund flows to investor expectations is greater for

funds with higher managerial skill uncertainty.

8 Consistent with the Bayesian approach, investors form their posterior expectations about managerial skill by combining their initial prior and the signal embedded in the historical fund performance. Now, if the skill signal is precise and unambiguous (corresponds to low volatility in historical fund performance), then the investors infer superior managerial skill, but in cases where the signal is noisy (corresponds to high volatility in historical fund performance), it is more likely that investors infer the lack of superior stock-picking skill in the fund manager.

28

Our second empirical prediction relates to the second source of imprecision: poor

quality fund information. To measure the quality of fund-specific information, we use

two fund characteristics as proxies: size of the fund family and size of the fund. We use

fund family size, measured by the total net assets under management of a fund family

each quarter, to proxy for information uncertainty due to poor information. Following

Sirri and Tufano (1998), we argue that smaller fund families are more likely to have

higher fund information uncertainty because they tend to have poor visibility and lower

brand awareness compared to larger families as well as high information disclosure costs.

In addition, the number of fund analyst and amount of media coverage for smaller fund

families is also quite low. As a result, potential mutual fund investors face higher search

costs and may find it difficult to evaluate fund quality based solely on fund- specific

information. Based on this, we predict that when investors face a dilemma when

choosing to invest in funds that belong to small and lesser known fund families, they put

more weight on their economic expectations compared to historical fund-specific signals.

Our second proxy for poor fund-specific information is fund size, measured by

total assets under management each quarter. We argue that investors perceive fund

information related to the quality of small funds with greater uncertainty. This is largely

due to three possible reasons: (i) extreme difficulty in differentiating themselves from

other existing mutual funds due to smaller advertising budgets, (ii) lack of credible or

verifiable historical information and (iii) possible lack of genuine skill. In addition, small

funds are highly risky investment partly due to unlimited growth opportunities with an

equally high probability of disappearance, as noted by Brown (1995). Whichever the

reason, we argue that investors are extremely reluctant to base their investment decisions

in small funds solely on fund-specific information. As a result of this reluctance,

investors rely on their expectations about future economic conditions in addition to fund-

29

specific information to make subsequent fund purchase decisions. Combining fund and

family size proxies, we formally state our second hypothesis:

Prediction 2: The sensitivity of fund flows to investor expectations is greater for

funds that belong to smaller fund families and funds that are smaller in size.

2.3 Data

2.3.1 Investor Expectations

To investigate whether investors' optimism affects their future mutual fund

investment decisions, we need a robust empirical measure that explicitly captures

investors' expectations about future economic growth. To this end, we construct a new

index of investor expectations from the survey data of the University of Michigan Survey

Research Center (UMSRC) from 1970 to 2008. Prior to 1978, the UMSRC data was

available at a quarterly frequency at 2, 5, 8, and 11 months, but from the beginning of

1978, the data is available at monthly frequency. For our purposes, we are interested in

the quarterly data and so we convert the monthly data after 1978 in a quarterly frequency

by including observations corresponding to 3, 6, 9, and 12 months.

Each month the UMSRC telephonically interviews a representative sample of

approximately 500 individual households across the United States and asks over 50 core

questions broadly related to their personal financial situation, their expectations about

future economic conditions in the United States economy, and their propensity to

consume major household items. Of these 50 questions, we focus on the five questions

that relate to households' expectations about future economic conditions in the U.S.

economy including changes in their personal financial situation, changes in business

conditions, unemployment situation, general price levels and changes in the interest rates

30

over the next 12 months. Specifically, the five questions as they appear in the UMSRC

questionnaire are:

A3. "Now looking ahead -- do you think that a year from now you (and your

family living there) will be better off financially, or worse off, or just about the

same as now?"

A7. "And how about a year from now, do you expect that in the country as a

whole business conditions will be better, or worse than they are at present, or

just about the same?"

A10. "How about people out of work during the coming 12 months -- do you

think that there will be more unemployment than now, about the same, or less?"

A11. "No one can say for sure, but what do you think will happen to interest

rates for borrowing money during the next 12 months -- will they go up, stay the

same, or go down?"

A12. "During the next 12 months, do you think that prices in general will go up,

or go down, or stay where they are now?" and "By about what percent do you

expect prices to go up, on the average, during the next 12 months?"

The relative score for each question is then calculated as the percentage of favorable

minus the percentage of unfavorable responses, plus 100, rounded to the nearest whole

number (See Appendix B for details).

Using these relative scores, we form a composite investor expectations index that

captures individual investors' overall economic outlook for next the 12 months which

includes both their future personal financial well-being and future macroeconomic

conditions. We start by aggregating the relative scores of the five survey questions in

each quarter over the entire sample and dividing it by the sum of the relative values of

31

the first quarter of the base year (1985).9 The choice of base year is motivated by the fact

that 1985 is neither a peak nor trough of a business cycle. The resulting index is defined

as the investor expectations index (INVEXP). Figure 1 shows the evolution of INVEXP

through historical business cycle peaks and troughs.

Panel A in Table 1 summarizes INVEXP along with each of its component

survey questions. We find that over the entire sample, the mean value of INVEXP index

is around 90, while the standard deviation is 11.03. We also find that the autocorrelation

coefficient declines exponentially from 0.79 in the first lag to 0.39 in the fourth lag. This

clearly indicates persistence in investors' economic expectations over time. This

persistence is a desirable property because it signals that investors carefully build their

economic expectations over an extended period of time and not as a knee-jerk reaction to

short-term favorable or unfavorable economic news such as one day stock market

crashes. Along with summary statistics, we also report the Augmented Dicky-Fuller

(ADF) test (with a constant and 4 lags) of stationarity for levels of INVEXP and its

components. The ADF test statistic for INVEXP is -4.01 (with the critical value of -3.49

at 1%) and clearly rejects the null hypothesis of non-stationarity. Thus, we use levels in

the predictive regression framework. In addition, Figure 2 plots the relative scores of

each component and the index levels of INVEXP and the NBER recession periods over

time.

A natural question that arises is whether INVEXP is a valid proxy of investor

expectations. Under ideal circumstances, a valid investor expectations proxy should

exhibit three important characteristics: (i) be a representative direct survey response of

investors, (ii) be highly correlated with future economic conditions, and (iii) reflect

information that other well-established economic expectations proxies capture. Given

9 For robustness purposes, we try different base years and also try the equally-weighted average of the relative scores of all the questions. Our results remain unchanged.

32

that INVEXP is constructed from the direct survey responses of a representative sample

of individual households in the U.S., it satisfies the first characteristic by construction.

To test whether INVEXP is highly correlated with future economic conditions, we

analyze the pair-wise correlation coefficient of INVEXP with different state variables

which are well-known in the literature to predict future economic conditions. The first

row in Panel B of Table 1 reports contemporaneous correlation coefficients, while the

second row reports correlations between INVEXP at time t-1 and state variables at time t.

Focusing on the first row, we find that INVEXP has the highest (positive and statistically

significant) correlation of 0.50 with future term spread, a strong indicator of short-term

business cycles (see Fama and French (1989), Chen (1991)).

Similarly, we find that INVEXP is correlated with changes in short-term interest

rates (-0.20), growth in industrial production (0.23) and stock market returns (0.18).

Given this evidence, we conclude with some degree of confidence that INVEXP captures

with future economic conditions with expected signs and satisfies the second

characteristic.10 Now, we turn to the third characteristic -- the empirical validation of

INVEXP index. To validate INVEXP, we use the Conference Board Expectations Index

(CBEXP) as a reasonable alternative proxy which captures similar information but from

different respondents and is well recognized in practice. The most important validation

test is the correlation between INVEXP and CBEXP. If both these proxies capture

similar information, then they should be positively correlated to each other. We find that

10

Several papers confirm our results in a more formal econometric framework. See e.g., Carroll (2003) who shows that the consumer expectations survey data (from UMSRC) on future inflation and unemployment rate have highly statistically significant predictive power for the changes in future inflation and unemployment rate even after controlling for the lagged inflation and unemployment rate; Bram and Ludvigson (1998) and Ludvigson (2004) who show that consumer confidence surveys have strong predictive power for quarterly consumer expenditure growth. More recently, Lemmon and Portniaguina (2006) show that the consumer expectations survey data (from the University of Michigan) does particularly a good job of predicting future economic activity, especially in the post-1977 period.

33

INVEXP is highly positively correlated (0.42) with CBEXP, which indicates that the two

proxies share a strong common component.

But if CBEXP captures similar information, the question is why not use it as a

proxy for investor expectations instead? The answer lies in several important differences

between the two proxies. First, INVEXP is based on survey responses to five questions

covering personal financial situation, business conditions, inflation, interest rates and

unemployment. CBEXP, in comparison, is based on three questions that are limited to

business conditions, unemployment and total family income. Second, the survey

questions in INVEXP focus on the economic conditions in the country as a whole, while

CBEXP questions focus on the economic conditions in the respondent's area of

residence. And lastly, INVEXP measures expectations over a longer term (12 months)

compared to CBEXP which measures expectations over a shorter term (six months).

Third, INVEXP correlates highly with TERM (0.50) and dSIR (-0.20) which are strong

indicators of future economic conditions, whereas CBEXP has statistically zero

correlation with TERM and has a opposite sign with dSIR (0.32). Thus, keeping these

important differences in mind, INVEXP seems a more suitable measure for our analysis

than CBEXP. Nonetheless, for robustness purposes, we reexamine some of our results

using CBEXP and report them in Tables 4 and 6.

2.3.2 Mutual Fund Sample

Our mutual fund data are from the Center for Research in Security Prices (CRSP)

Survivorship Bias Free Database. The data includes quarterly information related to

funds' total net assets (TNAs), returns (net of expenses), operating expenses, front-end

loads, inception date and fund management of all open-end equity funds offered to

investors from January 1970 through December 2008. Consistent with prior mutual fund

studies, we primarily focus on domestic equity funds which broadly include aggressive

34

growth, growth, growth and income, value and long-term growth funds.11 We exclude

sector, international and balanced funds from our analysis. In addition, we treat different

share classes of the same fund as separate funds due to difference in their fund flows,

loads and fees structures. To minimize the impact of fund size and age on our results, we

only focus on funds that have an average TNA of greater than 10 million dollars annually

and are in existence for more than one year. Based on these data filters, our sample

covers 12,562 unique fund-entities over 156 quarters spanning the first quarter of 1970 to

the last quarter of 2008. Table 2 describes the cross-sectional characteristics of our

mutual fund dataset over the entire sample period and over four subsamples: the decades

of 1970, 1980, 1990 and 2000.

The primary reason behind the choice of this particular time period and cross-

section is to provide a unique and perhaps the most comprehensive setting to test and

analyze the influence of investors' economic expectations on their mutual fund

investment decisions. The length of the sample period from 1970 to 2008 maximizes the

coverage of periods with relatively high economic optimism and pessimism, including

the era of Reagan optimism (late 1970s to mid 1980s), the Technology bubble (late

1990s), the housing market bubble (early 2000s) and the recent period of economic

pessimism after the financial crisis (mid 2007 on- wards). Similarly, the cross-sectional

coverage of our mutual fund data closely resembles the actual investment opportunity set

that individual investors face in the real world. Hence, these characteristics ensure that

our results provide a more realistic assessment of investors' investment decisions.

11

We include domestic equity funds that meet the following investment objective classifications: Policy code: CS, Hedge, TFE, MF, Pfd; the Wiesenberger code: G, G-I, G-I-S, G-S, G-S-I, I, S-G, I-S, I-S-G, I-G-S, S-I-G, S-G-I, I-G, S-I, GCI, SCG, LTG, MCG, IEQ; the Strategic Insight code: AGG, GMC, GRI, GRO, ING, SCG; and the Lipper Code: CA, DL DSB, EI, EIEI, ELCC, EMN, G, GI, LCCE, LCGE, LCVE, LSE, MC, MCCE, MCGE, MCVE, MLCE, MLGE, MLVE, MR, SCCE, SCGE, SCVE, SESE, SG, SP, SPSP. In case, a particular fund has more than one objective code in a particular year, we choose the objective code with greater granularity.

35

To capture the unobservable investment decisions of investors, we use observable

mutual fund flows as a proxy. We define net fund flows (FLOWS) as net growth in the

total net assets of funds, as a percentage of their total net assets, adjusted for prior period

returns. FLOWS are also adjusted for any increase in total net assets due to mergers.

Formally, it is defined as

������, =�� �,���� �,��������,������� �,�

�� �,���, (1)

where TNAi,t is the total net assets of fund i, Ri,t is the return net of expenses, MGTNAi,t

is the increase in the total net assets of fund i due to merger. Since the exact timing of the

cash flows is unknown, we assume that new money flows in and out of each fund at the

end of each quarter. We also assume that all dividends and distributions are reinvested in

the fund at the end of each quarter. FLOWS reflect how investors select different funds

to allocate their wealth based on their anticipation of future performance and economic

conditions. The first row in Table 2 reports the averages of the cross-sectional means and

standard deviations of FLOWS, measured at the end of each quarter. We find that, on

average, the equity funds have enjoyed relatively higher quarterly inflow of 18.22% in

the 1990s compared to the 1970s when the mutual fund industry was in its infancy.

Interestingly, even the cross-sectional dispersion of fund flows increased from 13.71% in

1970s to 63.25% in 2000s along with the size (both in terms of assets under management

and number of funds) of the mutual fund industry.

Now, we focus on the fund-specific characteristics that we need to control for and

will be used as independent variables in our analysis. These characteristics are well

known in the literature to be salient to fund investors and to affect individual mutual fund

flows. These typically include fund's historical performance, size, age, size of fund

family and total expenses.

36

First, we measure the fund's historical performance by the relative rank of the

fund's raw return (net of expenses) in each quarter. Each fund is assigned a relative rank

ranging from zero (worst performer) to one (best performer) each quarter. This

performance measure is in the spirit of Sirri and Tufano (1998) among others, who

documented that investors are "performance-chasers" and are highly sensitive to fund's

historical performance.

Second, fund size and fund age are defined as the total net assets under fund's

management at the end of quarter t and the difference between the fund's inception date

and the current quarter t, in years, respectively. The third and fourth rows in Table 2

describe the evolution of fund size and age over the last three decades. With the increase

in the size of the mutual fund industry, the average fund size increased almost four-fold

from $134 million in the 1970s to $495 million in the 2000s. Similarly, as the number of

funds offered to investors increased, the average fund age decreased from 12.4 years in

the 1970s to 7.32 years in the 2000s.

Third, we measure the size of the fund family as the sum of the total net assets

under management of all funds belonging to the same fund family in quarter t. We find

that the average size of a fund family over the entire sample is $13 billion. The average

fund family size increases from $12 billion in the 1980s to almost $43 billion in the

2000s.

Consistent with Sirri and Tufano (1998) and Barber, Odean and Zheng (2005)

who document the negative impact of total expenses on individual mutual fund flows, we

define total fund expenses as the expense ratio plus the up-front load amortized over a

seven-year holding period.12 The last row in Table 2 reports the annualized total expenses

over the last three decades. We find that the average total expenses across funds have 12 Because the expense data is available on an annual basis, we match the annual expense data with quarterly frequency without adjusting for the frequency change.

37

been monotonically decreasing from 1.82% in the 1970s to 1.44% in the 1990s and

finally to 1.32% in the 2000s.

In contrast, the average cross-sectional dispersion among funds' total expenses

has increased from 0.67% in 1970s to 0.87% in 2000s. We also consider the volatility of

funds' returns, measured by standard deviation of raw returns of funds over the past year,

as a fund-specific control variable. But the lack of strong and consistent empirical

evidence for the influence of fund return volatility over future fund flows discourages us

from using it as a control variable.13 Nonetheless for robustness, we control for fund

return volatility in our regression specifications, and in unreported results, find that our

findings remain robust. To minimize the influence of extreme observations on our

results, we winsorize fund flows, fund size and fund family data at the 1 % and 99%

levels.

2.3.4 Macroeconomic Control Variables

One might argue that individual investors' future fund purchase decisions are

primarily based on the most recent performance of equity mutual funds, which might be

influenced by prevailing macroeconomic conditions. To address this argument, we

control for the current macroeconomic conditions using two sets of factors: financial and

real-economy. The economic data are collected from CRSP, DataStream and the Federal

Reserve (FRED) databases for the period from 1970 to 2008 on a quarterly basis.

Following Chen (1991), the real economy factors include changes in Industrial

Production (dIndProd), changes in Real Personal Disposable Income (dRPDI), change in

Real Personal Consumption Expenditures (dRPCE) and changes in Unemployment Rate

(dUNEMR). The financial factors include default spread (DEF), term spread (TERM),

13

For example, Sirri and Tufano (1998) find no evidence of fund's raw return volatility influencing subsequent period fund flows. In contrast, Barber, Odean and Zheng (2005) find some evidence of fund return volatility being important to future fund flows.

38

stock market return (SMRET) and changes in short-term interest rates (dSIR).

Consistent with previous studies, the default spread is defined as the difference

between the bond yields of Moody's Aaa- and Baa-rated corporate bonds, while the term

spread is the difference between the bond yields of 10 year long-term government bond

and the three-month Treasury bill. The stock market return is the value-weighted return

(including dividends) of the NYSE/AMEX/NASDAQ stock indexes on a quarterly basis.

The short-term interest rates are defined as the market rate on three-month Treasury bill

at the end of each quarter.

2.4 Empirical Analysis

In this section, we empirically test our research hypotheses stated in section II

using a panel regression framework, pooled over thirty-nine years. In subsection A, we

analyze whether investor expectations about future economic growth influences future

fund flows in the presence of different fund characteristics and existing macroeconomic

conditions. And in subsection B, we test the role of fund-specific information uncertainty

as a plausible explanation for why investor expectations might be influence future fund

flows.

2.4.1 Influence of Investor Expectations on Future Fund Flows

Given our discussion in the hypotheses development section, we expect to find a

positive relationship between investor expectations and future fund flows. To test this,

we estimate the following panel regression equation:

FLOWSi,t = a + bINVEXPt-1 + c1Perfi,t-1 + c2FundAgei,t-1 + c3FundSizei,t-1

+ c4TotalFeesi,t-1 + c5FamilySizei,t-1 + dCategoryFlowst-1 + Controlst-1 + ei,t,

(2)

39

where FLOWSi,t is the percentage growth in the assets of fund i over quarter t, INVEXPt-

1 is the investor expectations measure for quarter t-1, Perfi,t-1 is the relative rank of fund

i's raw return at time t-1, FundAgei,t-1 is the number of years since fund inception at time

t-1, FundSizei,t-1 is natural logarithm of fund i's total net assets at time t-1, TotalFeesi,t-1 is

the total fund fees (operating expenses plus one-seventh of front end load fees) at time t-

1, FamilySize i,t-1 is natural logarithm of total net assets under fund i's family at time t-1,

CategoryFlowst-1 is the net growth in total net assets of all funds that belong a particular

Strategic Insights investment objective category at t-1.14 Controlst-1 is the vector of

macroeconomic controls which includes financial and real economy variables at time t-1.

Financial variables include default spreads (Default), term spreads (TERM), value-

weighted CRSP stock market returns (SMRET), and changes in short-term (dSIR)

interest rates. Real-economy variables include log growth rates in industrial production

(dIndProd), real personal consumption expenditure (dRPCE), real personal disposable

income (dRPDI) and unemployment rate (dUNEMP). The error term in the regression is

denoted by ei,t. To address any concerns regarding potentially correlated errors within

fund and time dimension (e.g., Petersen (2009) and Thompson (2011)), we estimate the

standard errors by clustering at the fund and time dimensions. The coefficient of interest

is b from the regression equation (2). It captures the sensitivity of subsequent fund flows

to level of INVEXP. Based on our discussions in section II, we expect, ex ante, this

coefficient estimate to be positive and statistically significant. That is, when investors are

optimistic about the future economy, equity mutual funds receive higher inflows in the

subsequent period, in addition to flows due to important individual fund characteristics

such as past performance as well as existing macroeconomic conditions.

14 The Strategic Insights investment objective categories include Aggressive Growth (AGG), Growth and Income (GRI), Growth (GRO), Income and Growth (ING), Mid-caps (GMC) and Small-caps (SCG).

40

The results presented in Table 3 strongly support our first hypothesis. As

predicted, we find a positive and statistically significant relationship between investor

expectations and subsequent fund flows across different regression specifications.

Column (A) presents the results of the regression specification with investor expectations

as the sole explanatory variable. We find a positive and statistically significant b

coefficient estimate. For example, a one-standard deviation increase in investor

expectations increases subsequent fund flows by 7.06% (0.0016*11.03*4) per year. That

is, when investors have optimistic expectations about their financial well-being and the

overall economy in coming months, the equity mutual funds receive 7.06% higher flows,

after controlling for different fund characteristics. But one might argue that the higher

inflows might just be due to the popularity of a particular investment objective category

("hot style") or favorable current macroeconomic conditions. To address this concern, we

include a category flow variable along with two sets of macroeconomic variables --

financial and real economy -- that proxy for current macroeconomic conditions in our

regression specification in columns (B) and (C). We find that the sign and statistical

significance remains unchanged but the economic significance of the investor

expectations coefficient b in column (B) diminishes marginally to 5.7% even in the

presence of lagged category flow and financial variables. Column (C) reports regression

results after controlling for both financial and real economy factors. Unlike financial

variables, we find that all coefficient estimates related to real-economy variables are

statistically insignificant and hence, have little or no ability in predicting future equity

fund flows. But consistent with our previous results, we find that the economic

magnitude and statistical significance of our investor expectations coefficient b is robust

to the inclusion of both financial and real-economy variables that proxy for current

macroeconomic conditions.

41

Following Sirri and Tufano (1998), we include different fund characteristics in

our regression specification in column (D). The results show that investors not only

consider fund-specific characteristics when making their fund purchase decisions as

documented by prior research, but also consider their own expectations about future

economic conditions. Specifically, we find that, on average, when INVEXP increases by

one standard deviation, future equity fund flows increase by 6.6% (0.0015*11.03*4)

even after controlling for different fund characteristics. This is a new and economically

significant result that empirically supports our first hypothesis. Further, we investigate

the robustness of our results in the presence of objective category flows and existing

macroeconomic conditions. Columns (E) and (F) report the results with category flow

and financial variables and both sets of macroeconomic control variables, respectively.

Importantly, our results remain robust to inclusion of these control variables in column

(E) and (F).

In Table 4, we repeat our analysis from the previous table using alternative

expectations proxies. This analysis emphasizes the informational advantages of INVEXP

and helps to differentiate it from the existing expectations proxies. Columns (A) to (C)

report the results of regression using lagged UMEXP as a proxy for investor expectations

and columns (D) to (F) report results using lagged CBEXP. For brevity, we report only

the coefficients in front of investor expectations proxies, but control for different fund

characteristics and macroeconomic variables in our regression specifications. As

expected, we find that coefficient estimates for UMEXP are positive and statistically

significant. Even though, the magnitude of coefficients are smaller than that of INVEXP

after controlling for macroeconomic variables in Table 3. Similarly, coefficients that

relate CBEXP to fund flows are positive but statistically insignificant when we control

for existing macroeconomic conditions. These results, in addition to results presented in

42

Panel B of Table 1, clearly highlight the informational advantage that INVEXP has over

existing expectations proxies, particularly CBEXP, in predicting future fund flows.

2.4.2 Role of Information Uncertainty in Fund Purchase Decisions

Consistent with our second hypothesis, we argue that investor expectations play

an important role in mutual fund purchase decisions particularly in the case of hard-to-

value funds characterized by high information uncertainty. Because investors are unable

to evaluate future fund quality based on noisy and poor fund-specific information, they

put less weight on historical fund-specific information with high information uncertainty

and more weight on their economic expectations when making fund purchase decisions.

To examine the conditional effect of INVEXP on the subsequent flows of funds with

higher information uncertainty, we run the following panel regression as our base

specification

FLOWSi,t = a + b1FIU×INVEXPt-1 + b2INVEXPt-1 + b3FIU + c1Perfi,t-1

+ c2FundAgei,t-1 + c3FundSizei,t-1 + c4TotalFeesi,t-1 +c5FamilySizei,t-1

+ dCategoryFlowst-1+ Controlst-1 + ei,t, (3)

where FIU is the representative dummy variable for fund information uncertainty

proxies, INVEXPt-1 is the level of the investor expectations index at time t-1,

FIU×INVEXPt-1 captures the differential impact of investor expectations on flows into

funds with higher information uncertainty compared to funds with lower information

uncertainty. Our coefficient of interest is b1. Following the previous literature, we include

lagged fund-specific controls such as fund performance, fund age, fund size, total fees

and fund family size along with aggregate investment category flows. We also control for

macroeconomic conditions using Controlst-1 which is a vector of lagged financial and

43

real-economy control variables. We apply this base regression specification to test

empirical predictions discussed in Section II.

2.4.2.1 INVEXP and Managerial Skill Uncertainty

To test our first prediction, we use volatility in a fund's risk-adjusted performance

as a proxy for managerial skill uncertainty. Following Pastor and Stambaugh (2002), we

define managerial skill uncertainty as the variation in the fund manager's ability to

generate returns in excess of passive benchmark and non-benchmark assets. We measure

fund managerial skill based on the fund's performance in excess of two benchmarks: the

one-factor (CAPM) model (a passive benchmark asset) and the Carhart four-factor model

(passive non-benchmark assets). For robustness checks, we also condition our fund

performance specifications on the macroeconomic variables as in Zheng (1999). The

following are different fund performance measure specifications:

1. Fund return in excess of the one-factor model:

��, − ��, =��� + ��

��� , − ��,� +!�, (4)

2. Fund return in excess of the conditional one-factor model:

��, − ��, =���′ + ���

�′�� , − ��,� + �"�� #$�� ∗ �� , − ��,�& +!�, (5)

3. Fund return in excess of the Carhart four-factor model:

��, − ��, =��' + ��,���(

' �)�� + ��,*�+' �), + ��,-�.

' /)� + ��,�0�' )�) +!�,

(6)

4. Fund return in excess of the conditional Carhart four-factor model:

44

��, − ��, =��'′ + ���,���(

'′ �)�� + ���,*�+'′ �), + ���,-�.

'′ /)� + ���,�0�'′ )�) +

�"�,���(' ($�� ∗ �)��) + �"�,*�+

' ($�� ∗ �),) +

�"�,-�.' ($�� ∗ /)�) + �"�,�0�

' ($�� ∗ )�)) +!�, (7)

where zt-1 is the vector of lagged predetermined variables discussed in Ferson and Schadt

(1996)15 and β2i is the vector of factor loadings on the predetermined variables. By

incorporating these predetermined variables in these asset-pricing models, we control for

any variations in excess fund returns due to public information. Moreover, the choice of

such broad fund performance specifications is motivated by the lack of a definitive or

uncontroversial managerial skill measure and the urge to test the robustness of our

results.

To obtain a fund's risk-adjusted performance, we follow the methodology of Sirri

and Tufano (1998). We estimate the time-series regression (for both unconditional and

conditional models) of excess fund returns on excess market return for the CAPM model

and excess market return, size, value and momentum for Carhart four-factor model using

all available monthly data. Then, we estimate the expected return implied by the models

and take its difference with the observed excess fund return. This difference is defined as

fund's risk-adjusted performance and acts as a proxy for fund managerial skill.16

We define managerial skill uncertainty as the time-series standard deviation of

the fund's risk-adjusted performance over the past 12 months. Using the volatility of the

fund performance, we classify funds in the top quintile of fund performance volatility as

ones that signal higher managerial skill uncertainty. We construct a dummy variable,

15 These public information variables include: (1) the lagged level of the one-month Treasury bill yield, (2) the lagged dividend yield of the CRSP value-weighted NYSE and AMEX stock index, (3) the lagged default or quality spread, (4) the lagged term spread in the bond market.

16 For example, see Berk and Green (2004). We calculate the quarterly fund performance as the three- month cumulative excess returns each quarter.

45

MSU that equals one if fund i belongs to the top quintile of fund performance volatility

in quarter t-1 and zero otherwise. We interact this dummy variable with the continuous

investor expectations variable to capture the differential impact of INVEXP on flows into

funds with the most managerial skill uncertainty compared to flows into funds with

relatively less skill uncertainty.

The results in Table 5 provide strong support for our first empirical prediction.

Funds with greater uncertainty about managerial skills receive higher inflows when

investors are optimistic about future economic conditions. In Panel A, we define

managerial skill uncertainty (MSU) as the volatility of the CAPM-adjusted fund

performance. Columns (A) to (C) report the regression results relating to the

unconditional fund performance specification in equation (4). We find that the coefficient

estimate on the interaction term b1 is positive and statistically significant, after

controlling for fund-specific characteristics such as past performance and fees. This

implies that funds with higher managerial skill uncertainty are rewarded with an increase

of 8.82% in flows when the INVEXP increases by one standard deviation at time t-1,

ceteris paribus. This is an economically important result. It strongly suggests that

investors rely on their own economic expectations, in addition to fund-specific

information, to make their investment decisions in funds with uncertain managerial skill.

Regression specifications in columns (B) and (C) test whether our results are robust to

the inclusion of investment category flows and prevailing macroeconomic conditions.

We find that our coefficient of interest, b1, remains unchanged, in sign, statistical and

economic significance. Columns (D) to (F) report the regression results based on

conditional fund performance measure specification in equation (5). These results only

strengthen our argument. The coefficient estimate b1 remains positive and statistically

significant even after accounting for any variations in fund performance due to public

information. In addition, our results are robust to the inclusion of financial control

46

variables alone or in combination with real-economy control variables which have little

impact on investors' fund purchase decisions, particularly for funds with inconsistently

performing fund managers.

To test the robustness of our results, we repeat our analysis using alternative fund

performance measures. In Panel B, we measure managerial skill uncertainty using the

volatility of fund performance adjusted for Carhart four-factors. Pastor and Stambaugh

(2002) argue that managerial skills are better captured by the Carhart four-factor model

(passive non- benchmark portfolios) than by a single-factor model (passive benchmark

portfolio). Columns (A) to (C) report the results based on unconditional fund

performance, while columns (D) to (F) report those based on conditional fund

performance measured by Carhart four-factor model in equations (6) and (7),

respectively. Consistent with prior results, investors' expectations are positively related to

fund flows into high managerial skill uncertainty funds. The results are statistically

significant and remain robust to the inclusion of fund-specific and macroeconomic

controls.

We repeat our analysis using alternative expectations proxies in Table 6. If the

alternative proxies capture investors' optimism/pessimism about future economic growth

as effectively as INVEXP, we expect them to be equally important for future flows in

funds with high managerial skill uncertainty. To test this, we interact UMEXP and

CBEXP with the high managerial skill uncertainty dummy across fund performance

specifications used in Table 5. For brevity, we report only the coefficient in front of the

interaction term between alternative expectation measures and the high managerial skill

uncertainty dummy. We find the sign of the interaction term coefficient remains positive

across different specifications, but find no consistent evidence of either UMEXP and

CBEXP being important to future flows in high managerial skill uncertainty funds.

47

Overall, these results alleviate any concerns that one might have about the informational

advantage of INVEXP over other existing investor expectations proxies.

In summary, investors' economic expectations matter in their purchase decisions

of funds that fail to unambiguously signal their high-quality managerial skill. This

conclusion holds regardless of the measure of managerial skill, conditioning on public

information, fund-specific controls and existing macroeconomic conditions.

2.4.2.2 INVEXP and Poor Quality Fund Information

In this subsection, we test our second empirical prediction that relates poor-

quality fund-

specific information and subsequent fund flows. We use two different fund

characteristics as proxies for information uncertainty due to poor fund-specific

information. First, we use size of the fund family as a proxy for information uncertainty.

We categorize the fund families as small, if their total net assets (TNA) under

management (in millions) belong to the bottom quartile in each quarter. We then

construct a dummy variable, FFS, that takes value one if the fund i belongs to a small

fund family and zero otherwise. The interaction term (FFS×INVEXP) measures how

investor expectations affect fund flows into small-family funds in comparison to fund

flows into large-family funds in the subsequent quarter.

Consistent with our second prediction, the results in Column (A) of Table 7 show

that the conditional effect of investor expectations on small-family fund flows, captured

by the coefficient b1, is positive and statistically significant. In general, when individual

investors hold a favorable view of the future economic conditions, funds belonging to

small sized families are more likely to receive higher inflows, despite its ambiguous

fund-specific information. In addition, the coefficient of the interaction term, b1, remains

48

statistically significant in the presence of fund-specific characteristics such as past

relative performance, total fees, age, size and fund family size. For example, small-

family funds receive 8.82% more inflows per year when investor expectations increases

by one standard deviation, other things remaining constant. The second and third

columns in Table 7 report the same test as the first column, but with the inclusion of

category flows, financial and real-economy factors to control for current macroeconomic

conditions. The coefficient on the interaction term remains unchanged in sign and

statistical significance, however, its magnitude marginally decreases. Thus, we conclude

that the conditional effect of investor expectations on small-family fund flows is robust

to the inclusion of fund-specific and macroeconomic controls.

Second, we use fund size as a proxy for fund information uncertainty. We

categorize a fund as a small-sized fund if its total assets under management are less than

the median fund size for that quarter. We also construct a dummy variable, FSU, that

equals one if the fund is categorized as a small fund and zero otherwise. As before, we

include an interaction term (FSU×INVEXP) between the FSU variable and the lagged

INVEXP variable that captures the conditional effects of investor expectations on fund

flows into small funds. Ex-ante, we expect small funds to receive greater inflow when

investors are optimistic about future economic conditions.

The results in Table 8 strongly support our prediction. We find that the

coefficient of the interaction term b1 is positive and statistically significant across all

specifications. In column (A), we report the regression results of our redesigned base

specification along with fund-specific characteristics. The sign and statistical significance

of the interaction term coefficient has two implications: (i) that the inflows in small funds

are positively related to investor expectations and (ii) that the inflows in small funds are

more sensitive to investor expectations compared to the inflows in large funds.

Specifically, we find that, on average, when the investor expectations increases by one

49

standard deviation, the small fund flows in the subsequent period increases by 9.27% per

year compared to that of larger funds, after controlling for fund-specific characteristics.

This is interesting and intuitively appealing result. Investors underweigh ambiguous

fund-specific information signals when making fund purchase decisions when they are

optimistic about future economic conditions. We then re-estimate the base regression

specification in Columns (B) and (C) and include the category flow and prevailing

macroeconomic controls. We find that the sign and the statistical significance of the

interaction term coefficient b1 remain unchanged, but the magnitude of the coefficient

marginally diminishes to 7.06%, confirming that our results are not due to any spurious

relationship between fund flows and investor expectations.

2.5 Summary and Conclusion

Much of academic literature focuses on understanding what information is

important to investors in making their investment decisions. While majority of these

studies focus on asset-specific information and current macroeconomic conditions to

explain investment decisions, very little attention is given to investors' optimism about

the future economy. This is surprising given that are strong theoretical arguments

emphasizing the role of investors' optimism in their future investment decisions.

In this paper, we investigate the importance of investors' economic optimism in

investment decisions and provide a plausible explanation to how investor optimism

enters investors' investment decisions. We show that investors' optimism about future

economic growth positively influences future fund flows even after controlling for

different fund characteristics and existing macroeconomic conditions. This result

suggests that investors increase their allocation into riskier assets, above and beyond

what can be justified by different asset characteristics and existing economic conditions,

50

when they are optimistic about the future. Further, we find that investors' optimism

becomes particularly important to investors whenever the fund-specific information is

noisy and less reliable. In particular, we find that fund flows respond more strongly to

investor optimism in funds with highly volatile past performance, funds that belong to

smaller fund families and funds with smaller fund size. Overall, this result suggests that

in situations when investors are uncertain about the quality of the asset based on its noisy

asset-specific information, they tend to put more weight on their expectations about the

future.

Another equally important contribution of this paper is that it provides a plausible

mechanism through which investors' optimism enters their fund investment decisions.

Using different fund characteristics as proxies for information uncertainty, we show that

when investors face noisy and uninformative fund-specific information, they rely more

on their economics expectations to make their fund purchase decisions. We find this is

particularly the case for: (i) funds with highly volatile past performance (high managerial

skill uncertainty), (ii) funds that belong to relatively smaller fund families and (iii)

relatively smaller sized funds.

Overall, this paper provides a set of novel empirical results that shed new light on

the role of investor expectations in individuals' investment behavior and improves our

understanding of how individuals make their investment decisions, particularly under

high information uncertainty.

51

Appendix

A. Variable Definitions

Variables Descriptions

Flows Net growth in the total net assets (TNA) under management of fund i between quarter t and t-1, as a percentage of their total net assets (TNA) under management at t-1, adjusted for returns and increase in their total net assets due mergers between quarter t and t-1.

INVEXP Sum of the relative scores (RS) of all five survey questions [(A3) to (A12)] at the end of quarter t-1 divided by the relative scores total at the end of first quarter of the base year (1985).

MSU = 1 if fund i belongs to the top quintile of funds sorted on the volatility of fund manager's alpha in the quarter t-1. = 0 otherwise.

FFSU = 1 if fund i belongs to a fund family with total net assets (TNA) under management (in millions) in the bottom quartile at the end of quarter t-1. = 0 otherwise.

FSU = 1 if fund i's total net assets (TNA) under management are less than that of a median-sized fund at the end of quarter t-1. = 0 otherwise.

Performance Rank of fund i's raw return at the end of quarter t-1 across all equity funds available to investors at the end of quarter t-1.

Fund Size Natural logarithm of total net assets (TNA) under management (in millions) of fund i at the end of quarter t.

Family Size Natural logarithm of total net assets (TNA) under management (in millions) of all the funds in fund i's family at the end of quarter t-1.

Expenses Expense ratio plus the up-front load, if any, of fund i in the previous year amortized over a seven-year holding period.

Fund Age The number of years at the end of quarter t-1 since the fund i was first offered to public.

Category Flows Net growth in total net assets (TNA) under management of all funds that belong to a particular Strategic Insight (SI) investment objective category between quarter t-1 and t-2, as a percentage of their total net assets (TNA) under management at t-2.

52

B. Details of Investor Expectations Index

In this appendix section, we provide a detailed description of investor expectations index and its construction methodology.

Instead of focusing on one blanket question about how investors feel about the future economy, each question in our index categorically focuses on the individual household's view on the future prospects of different individual components that, in aggregate, constitute the overall economy. For example, the first question categorically asks the households about their view on their personal financial future which primarily depends on the stability of their personal future income (which partly includes income from their financial portfolios) relative to the general price level. The second question inquires about the households' view on the overall business condition in the next 12 months. One might argue that such a question is too general in nature and might include several possible economic and financial factors. But this is not the case. In this survey, the interviewers ask the respondents whether the reason for their optimistic or pessimistic expectations about overall business conditions, is because of favorable or unfavorable news about higher consumer demand, stock market conditions, easy credit conditions, trade deficits or political conditions such government elections. Similarly, the third question focuses on respondents view on change in unemployment rate in the country as a whole. That is, whether the number of people out of work will increase, decrease or remain the same in the next 12 months. The fourth question relates to possible changes to borrowing interest rates. Again, the question categorically focuses on borrowing interest rate which is most likely to affect individual households' consumption and saving decisions. The last question in our study focuses on change in general price level or inflation. The question also probes households on a possible average percentage change in inflation level. For our purpose, the response to the first question is enough in determining households' expectations about future inflation level.

Now, we focus on the index construction methodology. All the responses to each of the five survey questions mentioned above are categorized into one of the three categories: favorable, unfavorable or indifferent. For instance, if a household responds that a year from now she expects to be financially better off, this is categorized as favorable response. If a household responds that she expects to be financially worse off, this is categorized as unfavorable response and if her response is that she expects her financial condition to be the same, the response is categorized as indifferent. Once the responses are categorized, relative scores for each of the five questions are calculated. For relative score calculations, UMSRC only considers favorable and unfavorable responses and ignores the indifferent responses. The relative score is calculated as the percentage of favorable minus the percentage of unfavorable responses, plus 100, rounded to the nearest whole number. For example, if 65% of respondents are optimistic about their future financial situation compared to 35% who are pessimistic, then the relative score of that question will be 130 ((65-35)+100) for a particular period. It is important to emphasize that we simply follow the methodology used by the University of Michigan Survey Research Center in calculating relative scores.17 Then, we aggregate

17 For more information, please visit the following link (http://www.sca.isr.umich.edu/documents).

53

the relative scores of five questions and divide it by sum of relative scores in the first quarter of 1985 to form our investor expectations index.

54

Figure 1 Investor Expectations and Historical Business Cycles, 1970-2008.

This figure plots the levels of investor expectations (INVEXP) index along with NBER Recessions from 1970Q1 to 2008Q4. The investor expectations (INVEXP) index is simply the aggregate of the relative scores of the survey questions in each quarter over the entire sample and divided it by the sum of the relative values of the first quarter of the base year (1985). The red bars indicates the recession quarters classified by NBER's Business Cycle Dating Committee.

55

Figure 2 Components of Investor Expectations, 1970-2008.

This figure plots the relative scores of five individual expectations questions along with the levels of the investor expectations (INVEXP) index from 1970Q1 to 2008Q4. The relative score is calculated as the percentage of favorable minus the percentage of unfavorable responses of each expectation question, plus 100, rounded to the nearest whole number. Panel A shows the relative scores of the survey question related to investors' expectations about their personal financial well being in next 12 months. Panel B plots the relative scores of the survey question that captures investors' expectations about the overall business conditions in next 12 months. Panel C plots the relative scores of the survey question related to the expected general price level (or inflation) in next 12 months. Panel D shows the relative scores of the survey question related to expected unemployment rate in next 12 months. Panel E reports the relative scores to the survey question that asks investors about their expectations of borrowing interest rate in next 12 months. Finally, the investor expectations (INVEXP) index is simply the aggregate of the relative scores of the survey questions in each quarter over the entire sample and divided it by the sum of the relative values of the first quarter of the base year (1985). Panel F plots the levels of INVEXP index over the entire sample period. The bars in each panel indicates the NBER recession periods.

56

Table 1 Data Description And Validation of Investor Expectations (INVEXP), 1970-2008.

Panel A reports the descriptive statistics investor expectations index (INVEXP) and its five components. Statistics include mean, standard deviation, Augmented Dickey-Fuller (ADF) stationarity test statistic (with a constant and 4 lags) and autocorrelation coeffici0ents (up to 4 lags). INVEXP is calculated as the equally-weighted average of the relative scores (the percentage of favorable responses minus the percentage of unfavorable responses, plus 100) of all the five survey question components, with the base year of 1985. The survey components include PFEXP, the relative score of expected change in personal finances; BCEXP, the relative score of expected change in business conditions in the economy; INFEXP, the relative score of expected change in inflation; UEREXP, the relative score of expected change in unemployment; and IREXP, the relative score of expected change in borrowing interest rate. Panel B reports the correlation coefficients of INVEXP and other existing expectations proxies with contemporaneous state variables which predict future economic conditions. Existing expectations proxies include UMEXP, the University of Michigan Expectation Index, and CBEXP, the Conference Board Expectations index. The state variables include TERM, the difference in yields of 3-month T-bill and 10-year government bond; dSIR, the changes in short-term (3-month T-bill) interest rate; dIndProd, the log growth rate in industrial production and SMRET, the quarterly return on the CRSP value-weighted index. All data are quarterly. ***, ** and * represent 1%, 5% and 10% confidence levels, respectively.

Panel A: Investor Expectations and its Components

Mean Std. Min Max ADF Autocorrelations N

Dev

Lag 1 Lag 2 Lag 3 Lag 4

INVEXP 90.64 11.03 62.87 122.15 -4.01 0.79 0.65 0.54 0.39 156

PFEXP 121.90 10.06 92.00 140.00 -2.28 0.85 0.81 0.72 0.66 156

BCEXP 108.17 13.59 73.00 146.00 -4.22 0.79 0.59 0.43 0.29 156

INFEXP 80.50 17.50 23.00 112.00 -3.73 0.61 0.44 0.45 0.35 156

UEMEXP 69.82 25.72 37.00 126.00 -4.74 0.78 0.60 0.45 0.27 156

IREXP 49.26 15.66 26.00 139.00 -3.99 0.79 0.61 0.51 0.36 156

Panel B: Correlations with State Variables related to Future Economic Conditions

TERMt dSIRt dIndProdt SMRETt

INVEXPt UMEXPt INVEXPt 0.50*** -0.20** 0.23*** 0.18**

1

INVEXPt-1 0.51*** 0.04 0.40*** 0.01

UMEXPt 0.16** 0.15* 0.48*** 0.09

0.36** 1

UMEXPt-1 0.11 0.10 0.37*** -0.06

CBEXPt 0.09 0.32*** 0.65*** 0.09

0.42** 0.80***

CBEXPt-1 0.00 0.18** 0.46*** -0.06

57

Table 2 Mutual Fund Sample Description, 1970-2008.

This table summarizes the cross-sectional characteristics of mutual fund flows and other fund-specific characteristics over the entire sample period from 1970 to 2008. In first three columns, we report the number of observations, means and standard deviations of cross-sectional averages across all domestic equity mutual funds over the entire sample. Rest of the columns report the means and standard deviations of the cross-sectional averages across all domestic equity over last four decades of 1970s, 1980s, 1990s and 2000s. The fund-specific characteristics include Fund Returns, Fund Size, Fund Age, Family Size and Total Fees. Fund Returns are defined as the historical raw returns net of expenses with a quarterly investment horizon. Fund Size is the total net assets under management of a fund at the end of each quarter. Fund Age is the difference between fund's inception date and the current quarter time t, in years. Family Size is the sum of the total net assets of all the funds under the management of the fund family at the end of each quarter. Total Fees are defined as the expense ratio plus the up-front load amortized over a seven-year holding period. Total Fees are reported on the annual basis.

Full Sample

1970s

1980s

1990s

2000s

Obs. Mean Std.

Mean Std.

Mean Std.

Mean Std.

Mean Std.

Fund Flows (%) 359,294 9.57 42.29

0.44 13.71

5.75 32.10

18.22 62.19

14.34 63.25

Fund Return (%) 369,338 2.29 4.85

1.72 5.04

3.69 4.82

3.69 4.94

-0.21 4.58 Fund Size ($ millions)

361,564 309.75 1254.49

133.99 274.92

203.23 422.26

424.95 1731.61

495.37 2737.00

Fund Age (years) 366,107 12.62 12.40

16.09 12.94

17.10 14.82

9.44 12.58

7.32 8.93 Family Size ($ millions)

312,361 13017.33 33079.80

55.99 69.24

71.54 72.56

11964.16 32225.06

42973.21 107382.60

Total Expenses (% annual)

99,215 1.56 0.77

1.82 0.67

1.63 0.75

0.07 0.78

1.32 0.87

58

Table 3 Regression of Fund Flows on INVEXP, 1970-2008.

This table reports OLS coefficient estimates of six different specifications of the panel regression described in equation (2). The dependent variable is Flows which measures the net flows in fund i between time t-1 and t. The independent variables include INVEXP which measures investors' optimism/pessimism about the future economic conditions over the next 12 months at time t-1. The coefficient of INVEXP is multiplied by 100. Perf measures the relative rank of fund i's performance at time t-1, Fund Age measures the number of years since fund inception at time t-1, Fund Size is natural logarithm of fund i's total net assets at time t-1, FamilySize is natural logarithm of total net assets under fund i's family at time t-1, Total Fees is the total fund fees (operating expenses plus 1/7th of front end load fees) at time t-1. CategoryFlows is the total inflows at time t-1 into each SI investment objective category. Macroeconomic Controls includes Financial and Real-economy variables at time t-1. Financial variables include default spreads (DEF), term spreads (TERM), value weighted CRSP stock market returns (SMRET), changes in short-term interest rate (dSIR). Real-economy variables include growth in industrial production (dIndProd), real personal consumption expenditure (dRPCE), real personal disposable income (dRPDI) and unemployment rate (dUNEMR). The data spans the period of 1970Q1 to 2008Q4. Standard errors reported in parentheses below the coefficient estimates are clustered for time. ***,** and * represent 1%, 5% and 10% confidence levels, respectively.

Dependent Variable: Net individual fund flows, FLOWSi,t

(A) (B) (C) (D) (E) (F) INVEXPt-1 0.1638*** 0.1294*** 0.1489*** 0.1523*** 0.0949** 0.0912**

(0.0541) (0.0461) (0.0561) (0.0566) (0.0409) (0.0451)

Perfi,t-1 0.1324*** 0.1202*** 0.1203***

(0.0131) (0.0118) (0.0118)

Fund Agei,t-1 -0.0011*** -0.0008*** -0.0008***

(0.0002) (0.0002) (0.0002)

Fund Sizei,t-1 -0.0549*** -0.0569*** -0.0569***

(0.0023) (0.0024) (0.0024)

Total Feesi,t-1 -1.9744*** -2.2398*** -2.2444***

(0.6294) (0.7151) (0.7160)

Family Sizei,t-1 0.0153*** 0.0183*** 0.0183***

(0.0012) (0.0012) (0.0012)

Category Flowt-1 1.9907*** 1.9856***

1.4858*** 1.4740***

(0.1019) (0.1019) (0.1271) (0.1174)

Defaultt-1 -8.0498*** -7.9582***

-8.6670*** -7.7134***

(1.0821) (1.2939)

(1.5475) (1.8523)

Termt-1 -0.7830** -0.8388**

-0.2956 -0.2901

(0.3494) (0.3444)

(0.3446) (0.3401)

SMRett-1 -0.0084 -0.0149

0.0234 0.0180

(0.0468) (0.0449)

(0.0335) (0.0377)

dSIRt-1 -0.0147 -0.0138

-0.0242 -0.0339

(0.0188) (0.0221)

(0.0187) (0.0263)

dIndProdt-1 -0.3691

0.1998

(0.4012)

(0.3096)

dRPCEt-1 -0.0006

0.0001

(0.0018)

(0.0021)

dRPDIt-1 0.0025

0.0002

(0.0041)

(0.0042)

dUNEMPt-1 -0.0015

-0.0008

(0.0012)

(0.0011)

Clustering (Fund) No No No Yes Yes Yes Observations 359,294 359,294 359,294 293,587 293,587 293,587

59

Table 4

Regression of Fund Flows on Other Expectations Proxies, 1970-2008: Robustness Checks. This table presents robustness checks of regression specifications in Table 3. The dependent variable is Flows which measures the net flows in fund i between time t-1 and t. Other Expectations Proxies include UMEXP, the University of Michigan Expectations Index and CBEXP, the Conference Board Consumer Expectations Index. The coefficients of UMEXP and CBEXP are multiplied by 100. We control for fund characteristics which include Perf, the relative rank of fund i's performance at time t-1, FundAge, the number of years since fund inception at time t-1, FundSize, the natural logarithm of fund i's total net assets at time t-1, TotalFees, the total fund fees (operating expenses plus 1/7th of front end load fees) at time t-1, FamilySize, the natural logarithm of total net assets under fund i's family at time t-1 and CategoryFlows, the total inflows at time t-1 into each SI investment objective category. Financial variables include default spreads (DEF), term spreads (TERM), value-weighted CRSP stock market returns (SMRET), changes in short-term interest rate (dSIR). Real-economy variables include growth in industrial production (dIndProd), real personal consumption expenditure (dRPCE), real personal disposable income (dRPDI) and unemployment rate (dUNEMR). The data spans the period of 1970Q1 to 2008Q4. Standard errors reported in parentheses below the coefficient estimates are clustered for fund and time. ***,** and * represent 1%, 5% and 10% confidence levels, respectively.

Dependent Variable: Net Individual Fund Flows - FLOWSi,t

(A) (B) (C) (D) (E) (F)

UMEXPt-1 0.2046*** 0.0728** 0.0641**

(0.0365) (0.0291) (0.0319)

CBEXPt-1

0.1332*** 0.0360 0.0129

(0.0336) (0.0254) (0.0305)

Fund Characteristics Yes Yes Yes Yes Yes Yes

Financial Variables

Yes Yes

Yes Yes

Real-Economy Variables Yes

Yes

Clustering (Fund & Time)

Yes Yes Yes Yes Yes Yes

Observations 293,587 293,587 293,587 271,056 271,056 271,056

60

Table 5 Relationship Between INVEXP And Managerial Skill Uncertainty, 1970-2008.

This table reports panel regression coefficient estimates of the equation (3). The dependent variable is FLOWS which measures the net flows in fund i between time t-1 and t. The independent variables include MSU x INVEXP, the interaction term between funds with high fund manager skill uncertainty and INVEXP between t-2 and t-1. The coefficient of MSU x INVEXP is multiplied by 100. INVEXP, the level of investors' expectations about the future economic conditions at time t-1; MSU, which measures the uncertainty related to managerial skill as the volatility of the fund performance in excess of CAPM model (Panel A) and Carhart four-factor model (Panel B) over past 12 months and equals 1 for the funds with time-series standard deviation of fund performance in top quintile at time t-1 and zero otherwise. The fund performance in Panel A and B is defined as sum of the intercept from the time-series regression of fund returns on the CAPM and Carhart model and their residual over the entire sample. For robustness checks, we reestimate the fund performance conditionally using equation (5) based on Ferson and Schadt (1996). Fund-specific controls include Perf, the relative rank of fund i's performance at time t-1, FundAge, the number of years since the fund inception at time t-1, FundSize, the natural logarithm of fund i's total net assets at time t-1, TotalFees, the total fund fees at time t-1, FamilySize, the natural logarithm of TNA under fund i's family at time t-1. CategoryFlows is the total inflows at time t-1 into each SI investment objective category. Financial variables include default spreads, term spreads, value-weighted CRSP stock market returns, changes in short-term interest rate. Real economy variables include growth in industrial production, real personal consumption expenditure, real personal disposable income and unemployment rate. The data spans from 1970Q1 to 2008Q4. Standard errors reported in parentheses below are clustered for fund and time. ***,** and * represent 1%, 5% and 10% confidence levels, respectively.

Panel A: CAPM Adjusted Performance Unconditional Model Conditional Model

(A) (B) (C) (D) (E) (F)

MSU x INVEXPt-1 0.0618** 0.0598** 0.0599** 0.0772** 0.0730** 0.0732**

(0.0271) (0.0266) (0.0267) (0.0326) (0.0316) (0.0317)

INVEXPt-1 0.0014** 0.0009** 0.0008* 0.0014** 0.0008** 0.0008*

(0.0006) (0.0004) (0.0005) (0.0006) (0.0004) (0.0005)

MSU -0.0838*** -0.0763*** -0.0763*** -0.1003*** -0.0926*** -0.0927***

(0.0264) (0.0253) (0.0254) (0.0286) (0.0277) (0.0277)

Perfi,t-1 0.1321*** 0.1201*** 0.1202*** 0.1320*** 0.1200*** 0.1201***

(0.0130) (0.0118) (0.0118) (0.0130) (0.0118) (0.0118)

Fund Agei,t-1 -0.0008*** -0.0005** -0.0005** -0.0008*** -0.0005** -0.0005**

(0.0002) (0.0002) (0.0002) (0.0002) (0.0002) (0.0002)

Fund Sizei,t-1 -0.0540*** -0.0561*** -0.0561*** -0.0539*** -0.0560*** -0.0560***

(0.0024) (0.0024) (0.0024) (0.0024) (0.0024) (0.0024)

Total Feesi,t-1 -1.9690*** -2.2334*** -2.2380*** -1.9701*** -2.2339*** -2.2386***

(0.6237) (0.7094) (0.7104) (0.6237) (0.7092) (0.7101)

Family Sizei,t-1 0.0151*** 0.0181*** 0.0181*** 0.0151*** 0.0180*** 0.0181***

(0.0012) (0.0012) (0.0012) (0.0012) (0.0012) (0.0012)

Category Flowt-1 1.4772*** 1.4650***

1.4767*** 1.4643***

(0.1268) (0.1171)

(0.1267) (0.1169)

Macroeconomic Controls

Financial Variablest-1

Yes Yes

Yes Yes

Real Economy Variablest-1

Yes

Yes

Clustering (Fund & Time)

Yes Yes Yes Yes Yes Yes

Observations 293,587 293,587 293,587 293,587 293,587 293,587

61

(Table 5 continued from previous page)

Panel B: Carhart 4-factor Adjusted Performance

Unconditional Model Conditional Model

(A) (B) (C) (D) (E) (F)

MSU x INVEXPt-1 0.0970*** 0.0931*** 0.0930*** 0.0773** 0.0726** 0.0727**

(0.0329) (0.0318) (0.0319) (0.0326) (0.0316) (0.0317)

INVEXPt-1 0.0013** 0.0008* 0.0008 0.0014** 0.0008** 0.0008*

(0.0006) (0.0004) (0.0005) (0.0006) (0.0004) (0.0005)

MSU -0.1062*** -0.0985*** -0.0984*** -0.1003*** -0.0923*** -0.0924***

(0.0290) (0.0281) (0.0282) (0.0286) (0.0278) (0.0278)

Perfi,t-1 0.1323*** 0.1202*** 0.1203*** 0.1320*** 0.1200*** 0.1201***

(0.0131) (0.0118) (0.0118) (0.0130) (0.0118) (0.0118)

Fund Agei,t-1 -0.0008*** -0.0006** -0.0006** -0.0008*** -0.0005** -0.0005**

(0.0002) (0.0002) (0.0002) (0.0002) (0.0002) (0.0002)

Fund Sizei,t-1 -0.0543*** -0.0564*** -0.0564*** -0.0539*** -0.0560*** -0.0560***

(0.0024) (0.0024) (0.0024) (0.0024) (0.0024) (0.0024)

Total Feesi,t-1 -1.9669*** -2.2325*** -2.2371*** -1.9700*** -2.2339*** -2.2386***

(0.6257) (0.7115) (0.7124) (0.6237) (0.7092) (0.7101)

Family Sizei,t-1 0.0152*** 0.0182*** 0.0182*** 0.0151*** 0.0180*** 0.0181***

(0.0012) (0.0012) (0.0012) (0.0012) (0.0012) (0.0012)

Category Flowt-1 1.4794*** 1.4672***

1.4768*** 1.4643***

(0.1270) (0.1172)

(0.1267) (0.1169)

Macroeconomic Controls

Financial Variablest-1

Yes Yes

Yes Yes

Real Economy Variablest-1

Yes

Yes

Clustering (Fund & Time)

Yes Yes Yes Yes Yes Yes

Observations 293,587 293,587 293,587 293,587 293,587 293,587

62

Table 6 Relationship Between Managerial Skill Uncertainty And Other Expectations Proxies, 1970-2008: Robustness Checks.

This table presents robustness checks of regression specifications in Table 5. The dependent variable is FLOWS which measures the net flows in fund i between time t-1 and t. Other Expectations Proxies include UMEXP, the University of Michigan Expectations Index and CBEXP, the Conference Board Consumer Expectations Index. The variables of interest are the MSU x UMEXP and MSU x CBEXP which capture the relationship between funds with high fund manager skill uncertainty, UMEXP and CBEXP between t-2 and t-1. The interaction term coefficients are multiplied by 100. Panel A and B present results with unconditional and conditional fund performance adjusted for CAPM model and Carhart four factor model, respectively. Fund Characteristics include Perf, the relative rank of fund i's performance at time t-1, Fund Age, the number of years since the fund inception at time t-1, Fund Size, the natural logarithm of fund i's total net assets at time t-1, Total Fees, the total fund fees at time t-1, Family Size, the natural logarithm of total net assets under fund i's family at time t-1 and Category Flows, the total inflows at time t-1 into each SI investment objective category. Financial variables include default spreads, term spreads, value-weighted CRSP stock market returns, changes in short-term interest rate. Real-economy variables include growth in industrial production, real personal consumption expenditure, real personal disposable income and unemployment rate. The data spans from 1970Q1 to 2008Q4. Standard errors reported in parentheses below are clustered for fund and time. ***,** and * represent 1%, 5% and 10% confidence levels, respectively.

Panel A: CAPM Adjusted Performance Unconditional Model Conditional Model (A) (B) (C) (D)

(A) (B) (C) (D)

MSU x UMEXPt-1 0.0387 0.0418

0.0276 0.0337

(0.0293) (0.0279)

(0.0292) (0.0282)

MSU x CBEXPt-1 0.0296 0.0289

0.0171 0.0184

(0.0240) (0.0220)

(0.0200) (0.0200)

Main Effects Yes Yes Yes Yes

Yes Yes Yes Yes Fund Characteristics Yes Yes Yes Yes

Yes Yes Yes Yes

Financial Variables Yes

Yes

Yes

Yes

Real-Economy Variables

Yes

Yes

Yes

Yes Clustering (Fund & Time) Yes Yes Yes Yes

Yes Yes Yes Yes

Observations 293,587 293,587 271,056 271,056

293,587 293,587 271,056 271,056

Panel B: Carhart 4-factor Adjusted Performance Unconditional Model Conditional Model (A) (B) (C) (D)

(A) (B) (C) (D)

MSU x UMEXPt-1 0.0507* 0.0607*

0.0258 0.0327

(0.0290) (0.0289)

(0.0298) (0.0280)

MSU x CBEXPt-1 0.0339 0.0375*

0.0162 0.018

(0.0220) (0.0210)

(0.0200) (0.0200)

Main Effects Yes Yes Yes Yes

Yes Yes Yes Yes Fund Characteristics Yes Yes Yes Yes

Yes Yes Yes Yes

Financial Variables

Yes

Yes

Yes

Yes Real-Economy Variables

Yes

Yes

Yes

Yes

Clustering (Fund & Time) Yes Yes Yes Yes

Yes Yes Yes Yes Observations 293,587 293,587 271,056 271,056

293,587 293,587 271,056 271,056

63

Table 7 Relationship Between INVEXP And Smaller Fund Families, 1970-2008.

This table reports panel regression coefficient estimates of equation (3) where FLOWS is the dependent variable and measures the net flows in fund i between time t-1 and t, FFS x INVEXP is the interaction term between funds with smaller sized fund families and INVEXP between t-2 and t-1. The coefficient of FFS x INVEXP is multiplied by 100. INVEXP is the level of investor expectations about the future economic conditions at time t-1, FFS measures the information uncertainty related to funds that belong to relatively small fund families and is defined as a dummy variable that equals 1 if the fund belongs to a small fund family and zero otherwise. We categorize the fund families as small, if their total net assets (TNA) under management (in millions) belong to the bottom quartile at time t-1. Perf measures the relative rank of fund i's performance at time t-1, FundAge measures the number of years since fund inception at time t-1, FundSize is natural logarithm of fund i's total net assets at time t-1, TotalFees is the total fund fees (operating expenses plus 1/7th of front end load fees) at time t-1, FamilySize is natural logarithm of total net assets under fund i's family at time t-1. CategoryFlows is the total inflows at time t-1 into each SI investment objective category. Financial variables include default spreads, term spreads, value-weighted CRSP stock market returns, changes in short-term interest rate. Real-economy variables include growth in industrial production, real personal consumption expenditure, real personal disposable income and unemployment rate. The data spans from 1970Q1 to 2008Q4. Standard errors reported in parentheses below are clustered for fund and time. ***,** and * represent 1%, 5% and 10% confidence levels, respectively.

Dependent Variable: Net Individual Fund Flows - FLOWSi,t

(A) (B) (C)

FFS x INVEXPt-1 0.0711** 0.0696** 0.0694**

(0.0304) (0.0299) (0.0300)

INVEXPt-1 0.0013** 0.0008* 0.0007

(0.0006) (0.0004) (0.0005)

FFS -0.0771*** -0.0619** -0.0614**

(0.0261) (0.0258) (0.0258)

Perfi,t-1 0.1324*** 0.1202*** 0.1203***

(0.0131) (0.0118) (0.0118)

Fund Agei,t-1 -0.0011*** -0.0008*** -0.0008***

(0.0002) (0.0002) (0.0002)

Fund Sizei,t-1 -0.0548*** -0.0569*** -0.0569***

(0.0023) (0.0024) (0.0024)

Total Feesi,t-1 -1.9841*** -2.2347*** -2.2391***

(0.6322) (0.7142) (0.7151)

Family Sizei,t-1 0.0137*** 0.0182*** 0.0183***

(0.0012) (0.0012) (0.0012)

Category Flowt-1 1.4851*** 1.4734***

(0.1272) (0.1175)

Macroeconomic Controls

Financial Variablest-1 Yes Yes

Real Economy Variablest-1 Yes

Clustering (Fund & Time) Yes Yes Yes Observations 293,587 293,587 293,587

64

Table 8 Relationship Between INVEXP And Smaller Funds, 1970-2008.

This table reports panel regression coefficient estimates of equation (3) where FLOWS is the dependent variable and measures the net flows in fund i between time t-1 and t, FSU x INVEXP is the interaction term between funds with smaller sized funds and INVEXP between t-2 and t-1. The coefficient of FSU x INVEXP is multiplied by 100. INVEXP is the level of investors' expectations about the future economic conditions at time t-1, FSU measures the information uncertainty related to funds that are relatively small in size and is defined as a dummy variable that equals 1 if the fund's total net assets under management are less than that of median fund size at time t-1 and zero otherwise. Perf measures the relative rank of fund i's performance at time t-1, FundAge measures the number of years since fund inception at time t-1, FundSize is natural logarithm of fund i's total net assets at time t-1, TotalFees is the total fund fees (operating expenses plus one-seventh of front end load fees) at time t-1, FamilySize is natural logarithm of total net assets under fund i's family at time t-1. CategoryFlows is the total inflows at time t-1 into each SI investment objective category. Financial variables include default spreads, term spreads, value-weighted CRSP stock market returns, changes in short-term interest rates. Real-economy variables include growth in industrial production, real personal consumption expenditure, real personal disposable income and unemployment rate. The data spans from 1970Q1 to 2008Q4. Standard errors reported in parentheses below are clustered for fund and time . ***,** and * represent 1%, 5% and 10% confidence levels, respectively.

Dependent Variable: Net Individual Fund Flows - FLOWSi,t

(A) (B) (C)

FSU x INVEXPt-1 0.1400** 0.1354** 0.1345** (0.0575) (0.0566) (0.0567)

INVEXPt-1 0.0007** 0.0001 0.0001 (0.0004) (0.0005) (0.0005)

FSU -0.2016*** -0.2016*** -0.2007*** (0.0521) (0.0508) (0.0509)

Perfi,t-1 0.1316*** 0.1195*** 0.1195*** (0.0131) (0.0118) (0.0118)

Fund Agei,t-1 -0.0010*** -0.0007*** -0.0007*** (0.0002) (0.0002) (0.0002)

Fund Sizei,t-1 -0.0685*** -0.0713*** -0.0713*** (0.0030) (0.0029) (0.0029)

Total Feesi,t-1 -1.9353*** -2.2017*** -2.2052*** (0.6010) (0.6857) (0.6863)

Family Sizei,t-1 0.0151*** 0.0181*** 0.0181*** (0.0012) (0.0011) (0.0012)

Category Flowt-1 1.4789*** 1.4712***

(0.1289) (0.1200)

Macroeconomic Controls Financial Variablest-1

Yes Yes

Real Economy Variablest-1 Yes

Clustering (Fund & Time) Yes Yes Yes

Observations 293,587 293,587 293,587

65

Chapter 3

To Group or Not to Group?

Evidence from Mutual Funds

66

3.1 Introduction

There is a large body of theoretical and empirical studies across a variety of

disciplines that examines the benefits of group versus individual decision making. The

idea that a “group mind” is distinctly different from a single one was first put forward by

the social psychologist Le Bon (1896). There is wide experimental evidence that implies

inferior choices made within groups than among individuals resulting from extreme

decisions by a dominant player in a team or a reduction in critical thinking for the sake of

unanimity with other group members.18 In economics, the negative effect of group

decision making is often linked to possible productivity losses caused by free-riding by

some team members (e.g., see Alchian and Demsetz, 1972; Holmstrom, 1982; Rasmusen,

1987; Nalbantian and Schotter, 1997).

There is an alternative literature that highlights the benefits of decision making

process within groups. Sah and Stiglitz (1986, 1991) point out that the aggregate “group

opinion” is the average opinion of all group members. Sharpe (1981) shows that teams in

the portfolio management industry are able to achieve diversification of style and

judgment. Barry and Starks (1984) provide a theoretical setting suggesting that teams in

investment funds may reduce portfolio risk. Yet, very few empirical studies provide

support to the opinion and risk diversification theories of groups. Hamilton, Nickerson,

and Owan (2003) find that teams increase worker productivity, and that this increase is

more apparent among earliest team members, high-ability workers, and heterogeneous

18 See Wallach and Kogan (1965) and Stoner (1968) for the phenomenon known as “risky shifts,” Moscovici and Zavalloni (1969), Kerr (1992), and Sunstein (2002) for “group polarization” concept, and Janis (1982) for the notion of a “groupthink.” Bone, Hey, and Suckling, (1999) observe that groups have no more consistency in decision making than individuals. Barber, Heath, and Odean (2003) find that groups are more likely to purchase stocks than individuals for “good reasons” even though these reasons do not improve performance.

67

teams. Adams and Ferreira (2010) analyze individual and group bettors in iceberg break-

up betting and find that teams arrive to less extreme decisions than individuals.19

The goal of this paper is to examine the effect of teams on fund performance,

their risk-taking behavior and other fund characteristics using a large U.S. equity mutual

fund database. Our analysis has are three distinguishing features. First, we focus not only

on the examination of the individual/team split for fund performance but also on the

understanding of the value of an extra group member for the benefits of team

management. The intuition here is that any group work always leads to a tradeoff

between a larger intrinsic knowledge base of the group versus a difficulty in arriving at

optimal decisions, especially under time constraints, which are present in many job

occupancies, including the mutual fund industry. Second, we differentiate the team

impact on fund performance across geographic locations. The intuition here is that the

value of adding a new member to a team must to be higher in large cities where each

individual is more likely to bring to the group his/her unique knowledge, skills, and

networking ability. Third, we look into the relation between team member characteristics

and fund performance. The intuition here is that individual characteristics of team

members must impact team performance even when team size and location is the same.

The list of empirical finance studies that deal with group and individual decision

making is not very long. For example, Prather and Middleton (2002) find no evidence of

differences in fund performance between group and individual decision making, but they

deal with data sample with large survivorship bias. Chen, Hong, Huang, and Kubik,

(2004) find underperformance among team-managed funds, while Bar, Kempf, and

Ruenzi (2010) and Bliss, Porter, and Schwarz (2008) find that single- and team-managed

19 Other evidence in favor of team decision making is based only on studies on signaling games experiments, such as Bornstein and Yaniv (1998), Cooper and Kagel (2004), and others. Blinder and Morgan (2005), based on experiments simulating monetary policy decisions by central banks, show that groups achieve better outcomes than individuals. There is also some support for the “wisdom of a crowd” phenomenon advocated by Surowiecki (2005).

68

funds exhibit similar average performance but teams have lower risk-adjusted returns and

smaller portfolio risk than individual managers. However, both these studies use CRSP

data and do not account for manager characteristics. Massa, Reuter, and Zitzewitz (2010)

compare single, named, and anonymous team management practices. Kostovetsky and

Warner (2011) study manager turnover differences in equity mutual funds while

controlling for manager team size, but they do not examine fund performance issues

related to fund management structure.

Our data comes from Morningstar Direct and covers the period between January

1992 and December 2010. Some studies provide evidence of better and more precise

coverage of mutual funds by Morningstar than CRSP (e.g., see Elton, Gruber, and Blake,

2001; Massa, Reuter, and Zitzewitz, 2010; Karagiannidis, 2010). However, these papers

do not systematize the disparity in fund management structure reporting. Therefore, as a

first step, we highlight the discrepancies between CRSP and Morningstar data on

managerial structure of funds. We show that very often CRSP reports single-managed

funds while these funds are team-managed in Morningstar, and vice versa. The existence

of these differences, which in some years in excess of 20% of the overall sample of

named equity mutual funds, could cast certain doubts on the results of many recent

studies that use fund manager-specific information using CRSP data.20 Indeed, the

impact of a team on fund performance using an exactly matched sample between CRSP

and Morningstar is very different for the two datasets. With CRSP data teams have no or

negative contribution to risk-adjusted returns computed based on unconditional and

conditional versions of Carhart (1997) model, while with Morningstar data teams show

not only positive but also often significant addition to fund performance.

20 The non-inclusive list of recent studies that use CRSP data on fund management structure include Agarwal and Ma (2011), Bar, Kempf, and Ruenzi (2010), Chen, Hong, Huang, and Kubik (2004), Cici, (2011), Dass, Nanda, and Wang (2011), Deuskar, Pollet, Wang, and Zheng (2011), Han, Noe, and Rebello (2008), Kempf and Ruenzi (2007), and Nohel, Wang, and Zheng (2010).

69

Next, we examine the difference between team and single-managed funds across

various aspects of fund performance. We observe that on average funds which are team-

managed have higher risk adjusted returns than their single-managed counterparts. This

result holds steadily after accounting for a range of fund and manager characteristics and

is present across various fund investment objectives except those in aggressive growth

category. We then examine the relation between the size of a fund management team and

fund returns. We observe that this relation is non-linear. In particular, we find that three-

person teams are the largest contributors to fund performance relative to single-managed

funds. This result corroborates well with the notion of increasing problems of free-riding

and decreasing cooperation effectiveness in larger groups (e.g., Alchian and Demsetz,

1972; Holmstrom, 1982; Mueller, 2012).21 We also investigate the benefits of group

decision making across various locations. We split the sample into funds whose advisors

are located in six financial centers as defined in Christoffersen and Sarkissian (2009) and

those located in smaller cities and repeat our tests. We show that only funds located in

financial centers gain from team management, interpreting this result as highlighting the

importance of learning and information spillover effects in larger cities (e.g., see Jacobs

1969; Glaeser, 1999). Locating in financial centers helps individual members to bring

more diverse knowledge and skills (informational diversity) to their teams and is

consistent with diversification benefits arising from team work argued by Sharpe (1981).

In addition, we find among funds in financial centers that those with more heterogeneous

team members in terms of age and undergraduate institution (social category diversity)

underperform those with more homogeneous managers. These results are consistent with

potentially larger frictions and conflicts of interests associated with non-homogeneous

21 Laughlin, Hatch, Silver, and Boh (2006) find that three-person groups are necessary and sufficient to perform better than the best individuals on highly intellective problems.

70

groups, as emphasized in Jehn, Northcraft, and Neale (1999), and career concerns issues

in the mutual fund industry raised in Chevalier and Ellison (1999b).22

Finally, we analyze whether team-managed funds exhibit different risk-taking

behavior than single-managed ones and what fund characteristics are associated with

team management. We find little evidence that team-managed and single-managed funds

differ statistically in their exposure to total risk, market risk, and idiosyncratic risk.

However, the volatility of team-managed funds is larger in economic terms than their

single-managed counterparts. Subsequently, we observe that a substantial part of this

excess volatility among funds with multiple managers comes from their statistically

larger loadings on small and value stocks. We further show that team-managed funds

lead to substantially lower turnover, more than 12% annually with a full set of fund and

manager characteristic controls. This result implies less aggressive trading within groups

of portfolio managers and, therefore, provides additional support that teams lead to less

extreme behavior. Teams also help funds bring more money: we find positive and

significant link between team management and net fund flows reflecting a recent trend in

mutual fund industry to rely more on team-managed funds. Thus, our study shows that

group-decision making in mutual fund industry has sizable performance benefits, but the

extent of these benefits depends on team size and diversity, as well as fund location.

The rest of the article is organized as follows. Section 2 presents the motivation

for our analysis and hypotheses development. Section 3 describes the fund- and manager-

level data. In section 4, we compare managerial structures reported in CRSP and

Morningstar databases and then conduct preliminary tests on the importance of team

management for fund performance using the two data sources. Section 5 presents the

main empirical findings of our paper. Section 6 examines the differences between team-

22 In their experimental study, Jehn, Northcraft, and Neale (1999) find that informational diversity positively affects group performance, but diversity in social categories and values among team members reduce this effect substantially.

71

managed and single-managed funds in terms of various measures of fund risk and several

fund characteristics. Section 7 concludes.

3.2 Motivation and Hypotheses Development

There is widespread evidence nowadays from the industry that mutual funds

prefer moving towards team management. For example, below is an excerpt from the

December 2, 2011 Reuters report:

“Mutual fund star managers have gone the way of the vinyl record: They're cool to have, expensive to get, and sometimes, not the best quality. In their place, fund companies like Federated Investors, Eaton Vance and Invesco are moving in favor of a team-oriented approach. Even Fidelity Investments, home of one of the first star managers, Peter Lynch, has switched some funds to a team-managed approach. The move helps fund companies defend against poaching, protect their funds’ returns, and shield themselves from the level of outflows seen at competing firms after their high-profile stars have flamed out.” 23

Recent academic sources also document the same trend (e.g., Massa, Reuter, and

Zitzewitz, 2010). This evidence however stands in stark contrast with the results of

numerous experimental and empirical academic studies that have tried, but with little

success, to identify benefits of group work in various fields of social science, including

finance and economics. For instance, papers such as Chen, Hong, Huang, and Kubik,

(2004), Massa, Reuter, and Zitzewitz (2010), Bar, Kempf, and Ruenzi (2010) among

others find that team management in mutual funds provides no gains over single-

managed funds and even often leads to inferior performance. Some literature from

economics that finds beneficial impact of teams on productivity and more balanced

23 “Funds move away from star managers, favor teams,” by Jessica Toonkel, December 2, 2011, Thomson Reuters.

72

decision making, such as Hamilton, Nickerson, and Owan (2003) and Adams and

Ferreira (2009), are based on extremely limited data. Therefore, our main hypothesis,

similar to most of the aforementioned studies, states the rationale for the existence and

the spread of teamwork in fund management, namely:

H1. Fund performance is higher among team-managed funds.

Note that the value of group decision making may greatly depend on internal and

external factors. First, there are many studies that examine team performance as a

function of team size. For instance, research shows that larger teams may often perform

worse than small ones (e.g., see Thompson, 2003; Mueller, 2012). While the earlier

literature has no clear answer on the optimal number of people in a group (on average,

varies between five and ten), it is obvious that the ideal team size should depend on the

tasks performed by individuals within a group. It appears that the more diluted the tasks

are, the smaller should be the optimal group size. In this respect, Mueller (2012) argues

that if companies deal with various coordination and motivational issues, then any group

composed of five or more individuals will already see significant increases in

coordination costs within the group and diminishing motivation across members of the

group. Hence, we can state our first prediction as follows:

P1. Fund performance is non-linear in the number of team members.

Second, the value of an additional team member must be greater under those

conditions when each individual has a higher potential to enhance the overall knowledge

and resource base of the group. In the fund management industry in particular, skills,

knowledge as well as networking ability of each team member can be of great

importance to fund performance. Numerous studies have shown that those conditions are

more readily available in larger cities (e.g., see Jacobs 1969; Glaeser, 1999;

Christoffersen and Sarkissian, 2009). Indeed, larger cities, especially financial centers,

73

can provide positive externalities to portfolio managers including, but not limited to,

easier knowledge transfer, faster and more diverse business connections, and potential

access to private information. Therefore, we can now formulate our second prediction:

P2. Fund performance is higher among team-managed funds located in larger cities.

Third, it is clear that individual characteristics of team members are important for

team decision making and performance. In particular, there could be differences between

more homogeneous and less homogeneous teams. For example, if a team of mutual fund

managers includes a much more senior and experienced person then the probability of

other team members to conform to the decisions of that individual increases (e.g., see

Janis, 1982). Jehn, Northcraft, and Neale (1999) observe that while the information

heterogeneity among group members is very helpful to group performance, the social

category heterogeneity is not. Moreover, Chevalier and Ellison (1999b) point out that

fund managers have different incentives at various stages of their careers, and so they are

not likely to collaborate well within teams composed of members of various age groups.

Thus, we can now articulate our third prediction:

P3. Fund performance is higher among more homogeneous team-managed funds.

Finally, numerous studies compare individual and group decision making to the

level of risk. We follow the arguments in Sah and Stiglitz (1986, 1991), Sharpe (1981),

and Barry and Starks (1984) and assume that working in teams does not induce extreme

risk taking behavior among portfolio managers. Thus, our second hypothesis can be

stated as follows:

H2. Team-managed funds do not take excessive risk.

74

Note that given the inconclusiveness of previous studies on the impact of group decision

making on fund performance, it is imperative to deal with precise fund managerial

structure data. We specifically address this issue in the next section.

3.3 Data

3.3.1 Main Data Source

Our primary data source is Morningstar Direct (MS, henceforth), a relatively new

survivorship-bias free institutional research product offered by Morningstar, Inc. This

database provides one of the most comprehensive and in-depth coverage of open-ended

mutual funds across the globe, including the United States. Our sample covers actively

managed U.S. diversified domestic equity funds with the following investment

objectives: Aggressive Growth (includes Small Company), Growth, Growth & Income,

and Equity Income from 1992 to 2010. We exclude all sector funds from our analysis

because their portfolios are constrained to follow a particular industry and hence are not

diversified. We also exclude index funds because majority of these funds are not actively

managed. MS reports all data at the fund share class level, including the names of the

fund managers. However, different share classes of the same fund have identical

underlying portfolio with the same fund manager(s). This might lead us to multiple

counting of fund management information and bias our analysis. To avoid such biases,

we aggregate mutual fund share class level observations to one fund level observation

using a unique fund identifier in MS.

To determine whether a fund is sole-managed or team-managed at the end of a

calendar year, we use the detailed fund manager data which includes fund manager

names, the exact date a fund manager joins and leaves a particular fund. We classify a

fund as sole- or team-managed based on the number of fund managers with the fund at

75

the end of calendar year. When only one fund manager is named at the end of calendar

year, we classify that fund as sole-managed for that year. Similarly, when two or more

fund managers are named with the fund, we classify the fund as team-managed. We

remove all fund-years which have missing or anonymous fund manager names or tenure

dates from our sample.24 Our final sample covers 3,935 unique funds with 35,440

manager-fund-year observations.

3.3.2 Fund Characteristics

For each fund we obtain information on total net assets under management,

expense ratios, turnover ratios, fund inception date, and fund family name from MS. This

information helps us control for fund characteristics that are well known in the literature

to affect individual fund performance. These characteristics typically include fund size,

measured by the total net assets under management of the fund at the end of calendar

year; fund age, defined as the difference between the fund’s inception year and the

current year; expenses, measured by the annual net expense ratio of the fund; turnover,

measured by the turnover ratio of the fund; fund family size, measured by the total net

assets under management of the fund complex to which the fund belongs at the end of

calendar year; fund return volatility, measured by standard deviation of raw net returns of

funds over the past year. We also include net fund flows, defined as the net growth in the

total net assets of funds, as a percentage of their total net assets, adjusted for prior year

returns. To minimize the effect of outliers on our analysis, we winsorize expense ratios,

turnover and annual fund flow variables at 1% and 99% levels.

24 The proportion of blank or anonymous entries for fund manager information in our initial data sample is only 7%. This stark difference with the percentage of anonymous funds reported in Massa, Reuter, and Zitzewitz (2010), which was reaching 18% in some years is due to the fact that Morningstar Direct has filled in names of managers for almost all funds (retroactively) after 2006.

76

Christoffersen and Sarkissian (2009) show that fund managers located in

financial centers earn higher returns than their peers located in smaller towns. To control

this location effect, we obtain the location information of fund advisors from MS.

Following Christoffersen and Sarkissian (2009), we define the following six cities to be

financial centers: Boston, Chicago, Los Angeles, New York, Philadelphia, and San

Francisco. If the fund advisor company is headquartered within a 50-mile radius of any

of these six cities, we classify the fund as located in the financial center.

It is important to point that our location variable differs from the previous studies.

Instead of using the headquarter location of the fund company or fund sponsor (e.g.,

Christoffersen and Sarkissian, 2009), we use the headquarter location of the fund advisor

company. For majority of funds, the fund advisor and the fund sponsor (the company that

offers the mutual fund to public) might be the same company (Chen, Hong, and Kubik,

2011). But for few funds they might be different because these funds choose to outsource

their portfolio management to third-party fund advisor companies. By choosing the fund

advisor location, we make analysis immune to the possibility of any bias due to third-

party fund management outsourcing.

3.3.3 Fund Manager Characteristics

In any study that examines potential impact of group decision making on fund

performance, it is important to control for the influence of manager’s demographic

characteristics.25 The demographic information available to us includes the name(s) of

fund manager(s), the name(s) of all funds they currently manage and have managed in

the past, their start and end dates with those funds, all undergraduate and graduate

25 Unfortunately, this has not been the case in many papers which attempt to determine the impact of team management of fund performance (e.g., Bar, Kempf, and Ruenzi, 2011; Massa, Reuter, and Zitzewitz, 2010).

77

degrees received, the year in which the degrees were granted, and the name of degree-

granting institution. In addition, we also have a detailed biographical sketch for all fund

managers from MS. This sketch is provided to MS by the fund managers themselves

which includes their personal and past work experience details. Following Chevalier and

Ellison (1999a), we use these data to create four manager characteristics variables:

Manager Tenure, MBA dummy, Average SAT, and Manager Age.

Specifically, we define the manager tenure as the difference between the year

when a fund manager started as a portfolio manager for a given fund and the current

year. To create the MBA dummy variable, we use the graduate degree details of each

fund manager in our sample. We define the MBA dummy variable as one if the fund

manager received an MBA degree and zero otherwise. To construct the average SAT, we

closely follow the methodology of Chevalier and Ellison (1999a). First, we obtain the

name of the undergraduate institution for each fund manager. Then, we look for that

institution’s SAT score in the 23-rd edition of Lovejoy’s College Guide (see Straughn

and Straughn, 1995). Most schools report the upper and lower of median verbal and math

scores for incoming student in that year. To calculate the composite SAT score for a

given school, we simply add the average of the upper and lower bounds of the verbal

score to the average of the upper and lower bounds of the math score. In few cases,

schools choose to report ACT scores instead of SAT. In those cases, we convert the ACT

to an equivalent SAT using SAT-ACT concordance tables provided by the College

Board.26

The construction of fund manager age variable is not straightforward because

very few fund managers in our sample disclose their date of birth in their biographical

sketch. To overcome this problem, we again follow the methodology proposed by

26 For more a detailed description of the construction of average SAT score, please refer to Chevalier and Ellison (1999a).

78

Chevalier and Ellison (1999a). For managers who report their date of birth, we simply

take the difference between the year of their birth and the current year. For managers

who do not report their date of birth, we construct an approximate manager age variable

by assuming that each manager was 21 year old upon receiving their undergraduate

degree. The limited coverage of undergraduate degree year information does reduce our

sample size, but does not affect our analysis.

An important difference between Chevalier and Ellison (1999a) and our study is

that they focus only on single manager funds, while our study focuses on both single-

and team-managed funds. It is relatively straightforward to create manager characteristics

for single-managed funds. But it is somewhat problematic to create manager

characteristics for teams of fund managers. Ideally, one might be able to create team

characteristics based on detailed understanding of the contribution of each team member.

Unfortunately, we do not have any these data. To overcome this problem, we simply

assume equal contribution of each team member. Hence, manager characteristics for a

team, such as manager tenure, age and SAT scores will simply be the equally-weighted

average of manager tenure, age and SAT scores of each fund manager in the team,

respectively. For the MBA dummy variable in case of teams, we define it to be one if any

one of the team members has a MBA degree and zero otherwise.

3.3.4 Fund Performance Measures

For computing fund performance measures we use each fund’s monthly net fund

returns from MS. We use three different performance metrics: objective-adjusted returns,

OAR, unconditional four-factor alpha, α(4U), using Carhart (1997) model, and

conditional four-factor alpha, α(4C), following the application of Ferson and Schadt

(1996) framework to Carhart (1997) model. We define OAR as the difference between

79

the average monthly return (net-of-fees) of a fund in the year minus the mean fund

returns across all funds for a given fund investment objective and year. We estimate each

fund’s unconditional and conditional risk-adjusted alphas using the following two

equations:

tititititmiiti eUMDmHMLhSMBsrr ,,, +++++= βα , (1)

and

( ) ( ) ti

Term

ttm

Term

i

Tbill

ttm

Tbill

ititititmiiti eZrbZrbUMDmHMLhSMBsrr ,1,1,,, +×+×+++++= −−βα , (2)

respectively, where ri,t is the monthly net fund return less the risk-free rate (proxied by

the one-month U.S. T-bill rate), rm,t is the monthly U.S. excess market return (i.e., the

return on the CRSP Value-weighted NYSE/AMEX/Nasdaq composite index less the

one-month U.S. T-bill rate), while αi is the risk-adjusted return, unconditional in Eq. (1),

α(4U), and conditional in Eq. (2), α(4C). SMB, HML, and UMD are returns on the size,

book-to-market, and momentum portfolios, respectively.27 In equation (2), Tbill

tZ 1− and

Term

tZ 1− are the two lagged (demeaned) public information variables: the one-month U.S.

Treasury bill rate (T-bill) and the term-structure spread (Term), defined as the difference

in yields on the 10-year U.S. government bond and three-month U.S. T-bill.

Funds change the number of fund managers from year to year. Therefore, we

remove all fund-years that have less than 12 monthly fund return observations and

estimate the fund alphas using their prior twelve monthly returns. Although the 12-month

horizon gives us fewer data points for the estimation than we may want, we believe that

given the high frequency of fund manager turnover, the longer (greater than one year)

27 These data are from Ken French’s site, http://mba.tuck.dartmouth.edu/pages/faculty/ ken.french/data_library.html.

80

estimation horizons will introduce bias in our analysis by incorrectly attributing fund

performance to a certain type management structure.

3.3.5 Summary Statistics

First, in Figure 1, we show the evolution of mutual fund management structure

from 1992 to 2010. It depicts the percentage of single-managed and team-managed funds

along with the total number of funds in each year of our sample. The total number of

funds increased from around 750 in the beginning of the sample period to more than

2,000 by 2010, peaking in 2007 with close to 2,500 funds. Consistent with reports in

other studies (e.g., Massa, Reuter, and Zitzewitz, 2010), we can see that the proportion of

single-managed funds has dropped significantly in the last two decades from almost 70%

in 1992 to around 30% in 2010.

Table 1 shows the summary statistics of mutual funds by the fund management

structure, where the data on team-managed funds is divided into funds with two

managers, three managers, four managers, and five managers or more. Panel A reports

the distribution (number and proportion in percent) of single- and team-managed funds

for each year in our sample. While all team-managed funds have increased their presence

in the industry, multiple-manager funds (five and more) have experienced the largest

relative and absolute gains in representation, four-fold from 4% in 1992 to 16% in 2010.

However, the largest proportion of team-managed funds has been directed by two

managers throughout our sample period.

Panel B of Table 1 reports three measures of performance, OAR, α(4U), and

α(4C), for single and team-managed funds. It also contains information about the

difference test in mean performance measures between each group of team-managed

funds and single-managed funds. We can see that team-managed funds show better

81

objective- and especially risk-adjusted performance. For example, the difference in OAR

between two-manager and single-manager funds is 0.014 per month or about 17bp per

year, while that between four-manager and single-manager funds is almost 56bp per year,

although this result is statistically insignificant. However, both fund alphas show that

three-manager funds, and, to some extent, funds managed by five or more people gain

the most relative to funds managed by a single person. For three-manager funds, the

differences in α(4U) and α(4C) are 43bp and 47bp per year, respectively, and these

results are significant at the 5% level. For five-plus-manager funds, the positive and

significant difference is observed only with respect to the unconditional alpha measure.

Other team sizes are not associated with significant outperformance relative to single-

managed funds.

Panel C of Table 1 reports mutual fund characteristics other than performance

measures. These include fund volatility, total net assets (Find Size), fund age (Fund

Age), turnover, and expenses. Among other fund characteristics, the notable differences

across managerial structures include turnover and expenses. Both these measures

decrease with an increase in the number of fund managers (and expenses decrease

monotonically). In addition, fund size tends to increase with team size. There are no

obvious differences however in fund volatility and age.

Finally, Panel D of Table 1 reports fund manager characteristics for our five

managerial structure groups. We notice that the average tenure with the same fund is the

highest among single-managed funds and so are the average SAT scores. Not

surprisingly, funds with larger teams are more likely to have at least one manager with an

MBA degree. The average age of managers appears relatively stable across both single-

managed and team-managed funds.

82

3.4 Management Structure: CRSP versus Morningstar

3.4.1 Fund Management Structure Differences

First of all, we determine the accuracy of funds’ management structure

information by comparing our MS sample to the widely used CRSP Survivorship Bias

Free Mutual Fund Database (CRSP, henceforth). Like MS, the unit of observation in

CRSP is the fund share class and the fund tickers are uniquely assigned to share classes.

To avoid double counting of fund’s management structure, we aggregate the share class-

level information to fund level for each fund. We match each fund in our MS sample to

CRSP using individual fund tickers and date of inception. In cases where the fund ticker

information is missing, we use fund names along with their date of inception for

matching purposes. We carefully do this matching by hand because there are differences

in fund naming conventions in both MS and CRSP. MS only reports the most recent

name adopted by the fund whereas CRSP reports different names adopted by the fund

over its active life. To ensure the accuracy of our matching strategy, we double check

each matched fund by hand. At the end, we are able to match 92.78% of our MS sample

funds to CRSP (3,651 out of 3,935 funds) sample between 1992 and 2010.

We also classify CRSP sample into single- or team-managed funds. For each fund

in a given calendar year CRSP reports the name of the fund manager(s) under “Portfolio

Manager Name” (also known as “mgr_name”) variable. We classify a fund as sole-

managed when only one manager name is listed and classify as team-managed when two

or more managers (or phrases such as “Team Managed” and “Investment Committee”)

are listed. We remove funds from our sample that report the name of the fund company

or their advisor(s) under the manager name variable. In addition, we also remove fund-

year observations for which the manager name is not available. We end up with 29,918

manager-fund-year observations in CRSP that represents an 84.42% match with our main

MS sample.

83

The table below provides an example of mismatch between the two data sources.

This example includes AARP Growth and Income Fund (CRSP Fund No: 53; MS

Fundid: FSUSA004ZG).

# Fund Managers Fund Name (MS) Fund Name (CRSP) Year MS CRSP SEC AARP Growth & Income

AARP Growth & Income Fund 1992 3 3 -

AARP Growth & Income

AARP Growth Tr: Growth and Income Fund

1993 3 1 -

AARP Growth & Income

AARP Growth Tr: Growth and Income Fund

1994 3 1 -

AARP Growth & Income

AARP Growth Tr: Growth and Income Fund

1995 3 1 3

AARP Growth & Income

AARP Growth Tr: Growth and Income Fund

1996 4 3 4

AARP Growth & Income

AARP Growth Tr: Growth and Income Fund

1997 5 1 5

AARP Growth & Income

AARP Growth Tr: Growth and Income Fund

1998 4 1 4

AARP Growth & Income

AARP Growth Tr: AARP Growth and Income Fund

1999 2 2 2

The table compares the fund name as well as the number of fund managers that manage

the fund at the end of the each calendar year for both CRSP and MS. To test the accuracy

of fund manager information in both databases, we compare this information to the one

provided by the fund to the financial regulator, the Securities and Exchange Commission

(SEC), each year. We hand collect the fund manager information from the fund’s

Prospectuses and other filings available on SEC’s EDGAR database each year. To

determine the number of fund managers in the SEC database, we count the names of fund

managers listed in the SEC filings at the end of the calendar year.28 The first and second

28 Creating the number of fund manager variable based on SEC filings is somewhat involved. We start by hand-collecting the fund’s Prospectus (Form N-1A), Annual Report (Form N-30D), and Post-Effective Amendments (Forms POS AM, 497, 485APOS and 485BPOS) available on SEC’s EDGAR database each year. Funds are legally required to include the full name, title, length of service, and business experiences of the individuals, including each member of portfolio management team who are primarily responsible for the day-to-day management of the fund in these filings. In cases where funds employ large portfolio teams, SEC requires the fund to provide information on at least five members who share the most significant responsibility for the day-to-day management of the fund's portfolio, for example, the managers with the largest percentages of assets under management. Funds are also required to disclose any change in fund

84

columns report the name of the fund given in MS and CRSP, respectively. Columns 4-6

show the number of fund managers reported in MS, CRSP, and SEC databases in a given

year, respectively. The first three rows in the last column have missing values because

we were unable to find corresponding year’s SEC filings on EDGAR’s website. This

table shows the managerial structure reported by Morningstar is consistent with SEC, but

we cannot say the same thing about CRSP data.

Table 2 reports the full extent of a misspecification in management structure

between CRSP and MS datasets for each year in our sample. Column 2 reports the

number of matched funds. We see that the overlap in funds between the two databases is

large in every year of our sample and it roughly follows the same trend as the overall

number of funds in our sample reported in Table 2. Column 3 and 4 as well as 5 and 6

report the percent of single-managed and team-managed funds in CRSP and MS

databases, respectively. We can observe that for the whole of 1990s, especially in the

beginning of the sample period, CRSP reports much more single-managed funds than

MS. Towards the 2000s, the overall proportion of single- and team-managed funds

becomes similar between the two databases. Columns 7 to 12 report misspecification

statistics. Columns 7 and 8 show the number of funds and their proportion that is

identified as single-managed funds in CRSP but are team-managed in MS. Columns 9

and 10 show the opposite problem, that is, the number of funds and their proportion that

is identified as team-managed funds in CRSP but are single-managed in MS. Finally,

column 11 gives the total number of mis-specified funds, while the last column indicates

the percent of misspecification in the overall matched sample.

manager(s) and provide information about the new manager(s) under the Securities Act through these filings. Each of these filings contains a filing date, which refers to the date the information was made public, and an effectiveness date, which refers to the date the information took effect. We then sort these filings based on their effectiveness date for each calendar year. Lastly, to determine the number of fund manager(s) in the fund, we simply count the name(s) of the fund manager(s) listed in the last SEC filing at the end of the calendar year. Because of the difficulty of doing this exercise over our entire sample of fund-year observations, we only checked several randomly chosen funds on the consistency of their Morningstar managerial data with SEC filings.

85

Columns 7-12 of Table 2 easily show that the largest misspecification in

managerial structure reporting between the two databases occurs in the early part of the

sample. The total misspecification is higher than 20% of the matched sample for most of

the 1990s. However, even in the 2000s, when both CRSP and MS report about the same

proportion of single- and team-managed funds (see columns 4 and 6), there is still a

significant misreporting in fund management structure that never goes below 10% of the

sample. Note that the average management structure misspecification over the whole

sample period is almost 19%. Taking into account the fact that we were not able to match

about 16% of MS sample with CRSP database, the actual misspecification in the reports

on the number of managers between the two databases is in excess of 20% during the last

two decades. The range of misspecification in CRSP is 17% to 29% for single-managed

funds and 6% to 23% for team-managed funds. Thus, Table 2 illustrates that the extent of

differences in management structure reporting between CRSP and MS databases is very

large and is likely to have a direct impact on studies using CRSP data in analyzing the

impact of teams in mutual funds returns.

3.4.2 Fund Performance Differences

Now we proceed to comparing the effect of team management on mutual fund

performance using CRSP and MS data. The regression model that we deal with has the

following general form:

tititititioti eFEControlsMgrControlsFundTeamccPerf ,,3,21,1,1, __ +++++= − δδδ , (3)

where Perfi,t is one of our performance measures, Team is the dummy for multiple-

manager funds, Fund_Controlsi,t-1 and Mgr_Controlsi,t are the sets of fund- and manager-

specific characteristics, while FEi,t includes the year and fund investment objective fixed

86

effects. Our fund-level controls are lagged by one period to exclude the contemporaneous

effect that they may have on fund performance.

Table 3 reports the results of panel regression tests of our two risk-adjusted

returns, α(4U) and α(4C), computed in a similar way from CRSP and MS databases on a

large set of fund and manager characteristics. In this table we again use our matched

sample between the two databases. The independent variable of interest is Team, defined

as a dummy variable which equals one if the fund has two (or more) fund managers and

zero if it has only one fund manager at the end of calendar year. Most of other

independent variables are defined in Table 1. To reduce the influence of outliers, we take

the natural logs of fund size, fund age, and manager age. Variable Flows is the net

growth in total net assets under management of the fund over the past year. SAT score is

divided by 100. All fund-level controls are lagged by one period except fund age. All

regression specifications include time and investment objective fixed effects (FE), and

the standard errors are clustered by fund. Each regression model also reports the number

of observations and the adjusted R2.

Panel A of Table 3 shows full sample estimations. There are 18,437 fund-year

observations with fund controls alone, but this number drops to 10,982 after the inclusion

of manager characteristics. Columns 1-4 report the estimation output using CRSP data.

Columns 1 and 2 show the estimates for α(4U), without and with fund manager controls,

respectively, while columns 3 and 4 show the corresponding estimates for α(4C). We can

see that in all these regressions, the coefficient estimate on Team is negative but

statistically insignificant. This result could explain conclusions in many papers that use

CRSP data that team management does not add any positive value for fund performance

(e.g., see Chen, Hong, Huang, and Kubik, 2004; Bar, Kempf, and Ruenzi, 2011).

Columns 5-8 of report the estimation output using MS data. Again columns 1 and 2 show

the estimates for α(4U), while columns 3 and 4 for α(4C), again without and with fund

87

manager controls, respectively. Now, we see that the results are drastically different. The

coefficient on Team is consistently positive across all estimations and, even though is not

always significant, is also economically sizable at least after accounting for both fund

and manager characteristics. Moreover, at the bottom of the panel we also report the test

results of the hypothesis that slope coefficients on Team in the corresponding MS and

CRSP estimations are the same, Team (MS-CRSP) = 0. As one can see, the difference is

positive and statistically highly significant across all four regression specifications. In

economic terms, this difference is 43-48bp per year, depending on the type of alpha, for

the tests that are based on estimates from regression models with a full set of control

variables.

It is worthwhile to mention the estimation results related to our control variables.

In particular, note that the coefficient estimates and their statistical significance are very

consistent across both CRSP and MS, unlike the results on the Team dummy, and are in

line with results in previous studies. Among fund-level characteristics, we observe that

fund size and expenses have large detrimental effect on performance. These results are

similar to findings in many other papers.29 However, funds benefit when they are part of

a larger family, again consistent with earlier studies (Chen, Hong, Huang and Kubik,

2004; Pollet and Wilson, 2008). We also document persistency in our two risk-adjusted

performance measures. Finally, there is also some evidence (for α(4U)) that higher

turnover reduces subsequent returns. As for the manager characteristics, consistent with

Chevalier and Ellison (1999a) we find a positive and highly significant relation between

fund performance metrics and managers’ SAT scores and no relation to MBA degree. In

addition, our results confirm that fund returns are higher for more experienced managers

with longer tenures at their respective funds (e.g., see Christoffersen and Sarkissian,

29 For the relation between firm size and performance see Chen, Hong, Huang, and Kubik (2004); for the relation between firm expenses and performance see Jensen (1968), Elton, Gruber, Das, and Hlavka (1993), Carhart (1997) and others.

88

2009). Note finally that even though the inclusion of manager characteristics drastically

reduces the total number of fund-year observations, the adjusted R2 indicate that they

provide incremental explanatory power for fund returns and therefore are important for

proper decoupling of the team management effect from manager-specific variables.

Panel B of Table 3 shows sub-sample estimations with unconditional Carhart

alpha as the only dependent variable over two non-equal periods, 1992-1999 and 2000-

2010. This non-equal time period split is motivated by some of the well-known earlier

results on the importance of teams for mutual fund returns, such as Chen, Hong, Huang,

and Kubik (2004), who use CRSP data over the 1992-1999 period and do not find any

benefits for team management. Each specification controls for fund and manager

characteristics but, for the sake of convenience, we report only the coefficient on Team

dummy alongside with its respective p-values. The evidence in Panel A that using MS

data leads to significantly more positive impact of team management on fund

performance is present also in sub-sample estimations. The test that slopes on Team for

the respective MS and CRSP regressions are the same, that is, Team (MS-CRSP) = 0, is

rejected for all specifications.

Thus, Table 3 shows that large discrepancies in management structure records

between CRSP and MS databases can translate to significant differences in team

management impact on fund performance. Ceteris paribus, MS data is able to provide

much more support for the benefits of group decision making in the mutual fund

industry.

3.4.3 Additional Misspecification Issues in Management Structure

There are two additional implications of the misspecification in management

structure data in CRSP which are important. First, one can no longer rely on the start

dates of fund manager(s) provided in this database, particularly in cases where more than

89

one fund manager names are listed. The start date (also variable known as “mgr_dt”) in

CRSP corresponds to a unique fund manager entry and specifies the date the current

manager(s) took control and assumed responsibility of the fund. For entries that list one

fund manager these dates might be less problematic, but for entries that list two or more

fund managers these dates might lead to serious errors. By giving one start date for funds

with two or more fund managers, CRSP leads researchers to assume that these managers

joined the fund on the same date which might not be true in all cases. And this is exactly

what we find in MS data, where in almost all team-managed funds, different fund

managers join the fund on different dates. Second, because CRSP provides incomplete

information on the number of fund managers (as shown previously), one also cannot rely

on the name of fund manager(s) provided in this dataset. Particularly, studies on manager

turnover which use fund manager names from CRSP might be affected from this

misspecification.

3.5 Team Management and Fund Performance: Empirical Tests

Having established that a researcher, using MS data, is more likely to find

evidence of positive contribution of team work in fund management industry, we now

directly examine the extent of its impact by using our full MS sample. Note that the

sample that we use for the reminder of the paper is larger than the one used in the CRSP-

MS matching tests in Table 3. Our goal is to analyze the potential benefits of team

management for various fund and/or manager characteristics.

3.5.1 The Average Effect of Team Management

Table 4 reports the results of the tests on the impact on team management on our

three measures of fund performance, OAR, α(4U) and α(4C). We report test results with

90

net (expense-adjusted) returns in Panel A and gross (expense-unadjusted) returns in

Panel B. As in Table 3, all regression specifications include time and investment

objective fixed effects, and the standard errors are clustered by fund. We also indicate the

number of observations and the adjusted R2. Again, the variable of interest is the Team

dummy. Most of our controls are also similar to those in Table 3 with two exceptions.

First, given some controversy regarding the inclusion of lagged dependent variable in

panel tests, we no longer consider lagged performance measures as additional

independent variables.30 Second, given the evidence of funds returns may be different

across geographic locations (e.g., Coval and Moskowitz, 2001; Christoffersen and

Sarkissian, 2009), now include a dummy variable for financial centers (FC) which equals

one if the fund is in a financial center and zero otherwise.

In columns 1-3 of Table 4, the dependent variable is the objective-adjusted

returns. We report the results without and with fund-level and manager-level controls. In

Panel A, the Team dummy comes up positive in all three regressions and is significant at

the 10% level in the most comprehensive specification that controls for both fund and

manager characteristics. In this latter regression, the economic impact of team

management on objective-adjusted fund returns is close to 40bp per year. In columns 4-6,

the dependent variable is the four-factor alpha. In this case, in Panel A even without

controls, the impact of team management is positive and significant at the 5% level. After

adding fund-level variables, its significance drops slightly to 10%, but with the inclusion

of manager characteristics, the coefficient on Team becomes significant at the 5% level,

and its economic magnitude increases by about 50% relative to that in column 4. In

columns 7-9, the dependent variable is conditional alpha. In Panel A, the coefficient on

Team again is positive in all three specifications, and while it is a bit less significant in

the first two regressions relative to the corresponding output in columns 4 and 5, it is

30 See Maddala and Rao (1973) and Grubb and Symons (1987) among others.

91

again significant at the 5% level for the most comprehensive last regression specification.

In fact, the economic impact of team management on conditional alpha after accounting

for all fund and manager characteristics is 46bp per year. The slopes on most of the

control variables in line with those reported in Table 3.31 In Panel B of Table 4, we

generally see the same pattern as in Panel A. There is only a small reduction in economic

and statistical significance of the coefficient on Team dummy for each estimation vis-à-

vis the corresponding test in Panel A. As before, the impact of team management on fund

performance is the largest after controlling for both fund-level and manager

characteristics.

We have observed that on average funds with team management practices appear

to do better than single-managed funds. The next natural inquiry is to determine whether

teams benefit all type of funds, irrespective of their investment objective. If team-induced

performance gains are concentrated in a specific fund category, then the most likely

explanation for previous findings will be not so much related to the benefits that teams

brings to fund operations but rather to the characteristics of that single fund category.

Table 5 reports the results of our tests on the impact on team management separately for

each of the four fund investment objectives. We show the outcome of tests for two risk-

adjusted measures of fund performance, α(4U) and α(4C), and report the same set of

estimates as in Table 4. The characteristics of regression models are also the same as

before but they always include both fund- and manager-level controls.

Columns 1 and 2 of Table 5 show that team management virtually has no impact

on aggressive growth funds returns. This could be due to the fact that aggressive growth

funds are believed to be benefitting the most among other fund types from higher

turnover rates; therefore, coordinating frequent trading decisions among multiple team

31 Note that the primary difference in the statistical significance of Team dummy between Table 4 and Table 3 comes from the increased sample size (e.g., more than 6% in tests with all control variables) rather than small changes in the set of control variables.

92

members may become impeding for fund performance. Note also that aggressive growth

funds deal with more “soft,” not easily available information about stocks and, as Stein

(2002) argued, in these cases, single-manager structures may be preferable. This is not

however the case for other objective categories. As shown in columns 3-8, managerial

teams have economically and statistically significant, at least at 10% level, relation to

risk-adjusted returns in all six estimations but one, for α(4U) of growth funds.32 Even in

this case, the economic impact of team management is 37bp per year, while that for

growth & income and equity income funds approaches a whapping 1.00% per year.

Therefore, Table 5 illustrates that having funds managed by teams of managers benefits

most of fund categories.

3.5.2 The Effect of an Additional Team Member

Our previous analysis shows that on average team-managed funds perform better

than single-managed funds, and this result holds across most of fund investment

objectives. Clearly, another relevant question is whether the positive relation between

team management and fund returns is linear in team size. Prior research is very scarce on

this issue. One evidence of non-linear benefits of team size is present in Hamilton,

Nickerson, and Owan (2003), who find largest increases in productivity of garment

industry workers when they join the teams at the early stages of team formation. In an

experimental study, Laughlin, Hatch, Silver, and Boh (2006) find that when dealing with

highly intellective problems three-person groups are necessary and sufficient to perform

better than the best individuals, and that groups with more members do not add extra

performance gains.

32 Note that some drop in the statistical significance of Team dummy for growth and equity income funds simply occurs because of the reduction in sample size rather than from the decrease in the magnitude of coefficients from the full-sample estimation in Table 4.

93

Recall from our Table 1 (Panel B) that team size indeed appears to be important

to fund returns, and that the largest gains in risk-adjusted performance are observed

among funds administered by three managers. What is necessary to do now is to examine

if this pattern persists or changes after controlling for our usual sets of fund and manager

characteristics. Therefore, we run the following regression model:

titititi

tititititi

eFEControlsMgrControlsFund

FMcFMcFMcFMccPerf

,,3,21,1

,4,3,2,10,

__

5432

+++

+++++=

− δδδ, (4)

where 2FMi,t, is a dummy which equals one if the fund has two fund managers at the end

of calendar year and zero otherwise; 3FMi,t, is a dummy which equals one if the fund has

three fund managers at the end of calendar year and zero otherwise; 4FMi,t is a dummy

which equals one if the fund has four fund managers at the end of calendar year and zero

otherwise; and 5FMi,t is a dummy which equals one if the fund has five (or more) fund

managers at the end of calendar year and zero otherwise. The other variables are defined

as before.

Table 6 shows the estimation results of fund management team size on the two

measures of risk-adjusted fund performance, α(4U), and α(4C). Consistent with results

of simple difference tests in Panel B of Table 1, the three-manager funds add the most of

performance gains vis-à-vis single-managed funds in terms of both unconditional and

conditional alphas. The economic value of a three-person team management on fund

performance ranges between 50bp and 60bp per year for the specification that includes

all control characteristics (0.04 and 0.05 percent per month, respectively). Teams with

two managers as well as larger teams (four and five or more managers) add less

performance gains relative to single-managed funds. These extra benefits are not always

statistically significant even at the 10% level which is achieved only among funds with

five or more managers. However, note that not only statistical significance of funds

94

managed by five people but also their economic impact decreases with the inclusion of

control variables, especially after accounting for average manager characteristics. Having

said that, the economic value of team management for funds that are managed by two,

four, or five or more managers can still remain sizable, although always less than that for

funds managed by three people in each respective tests specification. For instance, for

funds with five or more managers the annual impact of team management on their

conditional alpha is 43bp, as reported in column 6 (it is 60bp for three-manager funds).

Thus, Table 6 confirms our prediction P1 and shows that team size is non-linearly

related to fund performance. Intuitively, the number of team members determines the

tradeoff associated with larger knowledge base that more people bring to the team versus

coordination costs among multiple individuals, as indicated by Mueller (2012) and

others. This result is also consistent with Hamilton, Nickerson, and Owan (2003). Each

group member brings his/her specific skills and talents, but large cohorts of people with

various views on the subject matter may reduce productivity due to higher difficulty of

arriving to unanimous conclusions.

3.5.3 Team Management and Geographic Location

If teams in the financial industry are able to achieve diversification of style and

judgment, as argued by Sharpe (1981), then the value of having a team must be more

profound under those conditions when there are more objective reasons for people in

groups to have “uncorrelated” to each other views. This can occur more easily in larger

cities than in smaller communities. Indeed, group members in larger cities may have

more independent sources of information and more diverse networking potential than

residents of small towns. Therefore, we test this idea by examining now the team impact

95

on fund performance in financial centers versus smaller towns. The regression model is

as follows,

titititi

iitititi

eFEControlsMgrControlsFund

FCcFCTeamcTeamccPerf

,,3,21,1

3,2,10,

__ +++

++×++=

− δδδ, (5)

where Teami,t×FCi is the interaction term between the dummies on team management

and financial center dummies.

Table 7 reports the estimation results of fund management team size on our two

risk-adjusted measures of fund performance, α(4U), and α(4C). Besides reporting the

usual outcome of estimations, for each regression it also shows the results of the F-test of

the hypothesis that the performance of team-managed and single-managed funds is the

same. These tests are conducted separately across funds whose advisors are located in six

financial centers and those outside that set of cities. Columns 1, 2, and 3 of the table

show the estimation results for the unconditional alpha without controls, with fund

controls only, and with full set of control variables, respectively. We can see that in all

three specifications, the coefficient on Team is statistically zero (sometimes positive,

sometimes negative), implying that teams add no gains to performance for funds not

located in financial centers. The F-test at the bottom of the table restates these results.

However, the value of a team is diametrically opposite in financial centers. First, the

coefficient on the interaction term is consistently positive and economically significant,

indicating extra benefits of team management in financial centers versus other places.

Second and more importantly, the F-test shows that in financial centers team-managed

funds always significantly (at the 5% level) outperform single managed funds.33

33 In these tests, we test whether the combined coefficient of the team impact on fund performance, c1+c2, is positive and statistically significant since both Team and FC are dummies and here take the value of unity.

96

Our estimations with conditional alpha in columns 4-6 of Table 7 lead to the

same findings. Again, we observe no gains to managing funds in teams if the locations of

funds advisors are outside financial centers. When funds are in financial centers, the

evidence of benefits of group-decision making is even higher than before. Both economic

and statistical results are stronger than in the case of unconditional alpha. For instance,

for the regression specification with the full set of control variables (column 6), the

marginal value of multiple manager funds versus single-managed ones is almost 70bp

per year, and this difference is statistically significant at the 1% level. All these findings

confirm our prediction P2.

The results in Table 7 support Sharpe (1981) arguments and provide novel

evidence that group decision making is more beneficial in such environments where

group members are more likely to acquire knowledge and skills and establish business

connections. Clearly, at least in the finance industry in general and mutual fund industry

in particular, this becomes more achievable in financial centers than in smaller cities. Our

evidence also highlights a new example of superior learning and/or knowledge spillover

effects in larger cities as argued by Jacobs (1969), Glaeser (1999) and others.

3.5.4 The Role of Team Diversity

Besides the tradeoff between group and individual decision making and the

determination of the optimal size of a team, the other important question is the potential

effect of group diversity on performance. The literature on diversity in teams has led to

inconclusive results regarding the impact of group composition on their performance (see

Williams and O’Reilly, 1998; Jehn, Northcraft, and Neale, 1999; Hamilton, Nickerson,

and Owan, 2003; Van Knippenberg and Schippers, 2007). On the positive side, larger

diversity in team members may enhance information processing skills of the group as a

whole; on the negative side – significant differences among team members may cause

97

frictions, conflicts of interests, and within-group biases. Most of the aforementioned

studies deal with limited experimental and empirical data.

Our rich mutual fund dataset with various characteristics of fund managers

provides an ideal testing ground for the examination of the effect of group diversity on

fund performance. In particular, we can create diversity proxies across three dimensions

of fund manager characteristics: tenure with the fund, SAT score, and age. As a diversity

measure we use the coefficient of variation. It is the ratio of the standard deviation of a

variable over its mean, and it is a useful statistic for data which can only take non-

negative values (e.g., see Allison, 1978). Thus, our diversity proxies are:

)(/)( ,,, tititi TenureTenureDiversityTenure µσ= , (6)

)(/)( ,,, tititi SATSATDiversitySAT µσ= , (7)

)(/)( ,,, tititi MAgeMAgeDiversityMAge µσ= , (8)

where σ and µ are the standard deviation and mean of the corresponding manager

characteristic, respectively. The table below reports the summary statistics of these

diversity measures.

Mean S.D. Min Max Median

Tenure Diversity 0.6313 0.3468 0.0338 2.0718 0.6082

SAT Diversity 0.0990 0.0574 0.0022 0.3735 0.0945

MAge Diversity 0.1834 0.1110 0.0111 0.6985 0.1746

All average and median diversity measures are within 0-1 range. The largest spread in

these measures is observed for the fund tenure diversity, the smallest for SAT score

diversity.

98

Table 8 shows the impact of team diversity on fund performance for funds

located in financial centers and other places. We immediately focus on geographic

breakup of our sample since we already determined a primary impact of team

management on funds located in larger cities. The table reports the estimates from panel

regressions of unconditional and conditional fund alphas on three team diversity

measures defined by Eqs. (6-8), the number of observations, and the adjusted R-

squares.34 Columns 1 to 4 show the results for funds in financial centers, while columns 5

to 8 – in other locations. In columns 1 and 3 financial center fund alphas are regressed

only on the three manager diversity measures with no any controls. We observe

significant economic and statistical impact of diversity in SAT scores and manager age

on fund performance, and this relation is negative. This implies that homogeneous teams

in financial centers outperform heterogeneous ones. The diversity in manager tenure does

not appear to play an important role for fund returns. After controlling for the full set of

fund and manager characteristics, including the team size in columns (2) and (4), our

earlier conclusions only strengthen. Now, the values and statistical significance of

coefficients on manager diversity measures based on SAT scores and age increase, while

retaining the negative sign. A one standard deviation (0.06) increase in the SAT score

diversity increases unconditional and conditional alphas by about 50bp and 70bp per

year, respectively, while one standard deviation (0.1) increase in manager age diversity

leads up to 60bp annual performance boost based on conditional alpha. We do not find

any consistent evidence for the importance of diversity in team members among funds

located outside financial centers, illustrating again the irrelevance of team management

for fund performance for these types of funds. The only significant outcome occurs with

34 Note that our sample size now is much lower than in the earlier tests. This drop occurs for the following two reasons. First, in the current tests we use only team-managed funds. Second, when only one manager in a team has identifiable characteristic, it is impossible to compute the diversity measure based on this characteristic. However, these observations still contribute to the sample that contains average manager characteristics.

99

manager tenure diversity which results in a positive slope when the dependent variable is

conditional alpha after controlling for fund and other manager characteristics.

Thus, our findings support other papers on team diversity that highlight more

problems than benefits associated with grouping people with different characteristics into

the same teams (e.g., Jehn, Northcraft, and Neale, 1999). The results are also consistent

with career concerns issues in mutual funds (e.g., see Chevalier and Ellison, 1999b).

Managers with large differences in incentives and career options, stemming from

differences in their educational background and age, are unlikely to collaborate well on

such vaguely defined issues as fund portfolio composition and trading activity.

3.6 Team Management, Risk Taking, and Fund Characteristics

After analyzing various aspects of performance differences between team-

managed and single-managed funds in the earlier part of the paper, in this section, we

examine whether there exist systematic differences in risk taking and other fund

characteristics that can be distinctly attributed to group decision making in mutual fund

industry. First, recall that the existing literature is unclear on the impact of team on risk

taking. Some studies, such as Wallach and Kogan (1965), Stoner (1968), Sunstein

(2002), and others find that groups could act more aggressively and undertake more risk.

Other studies, however, such as Barry and Starks (1984) and Adams and Ferreira (2009),

provide theoretical and some empirical evidence that groups, may in fact, reduce risk. To

address these issues within our framework, we use the following model:

titititititi eFEControlsMgrControlsFundTeamddRisk ,,3,21,1,10, __ +++++= − δδδ , (9)

where Riski,t is one of fund’s i risk measures at time t. We consider several risk measures.

The first is the total volatility of the fund. The second is market risk and the idiosyncratic

100

residual volatility coming from the standard CAPM. The final set comes from the

Carhart (1997) model (see Eq. (1)) and includes market beta, the loadings on size, book-

to-market and momentum portfolios, as well as the idiosyncratic residual volatility from

this model.

Table 9 reports the results of the estimation of the impact of team management on

various risk measures. In this table, the market and residual risk from the CAPM are

denoted by Mrk1 and IdVol1, respectively, while these risks from the Carhart (1997)

model as Mrk4 and IdVol4, respectively. Each regression specification includes a full set

of fund and manager controls as in previous tests with the exception of two fund-level

variables, namely, fund family size and net flows. There are no a priori expectations

about the impact of those two variables have on risk characteristics of funds. We can see

that team management has no statistically significant impact on funds’ total risk, market

risk, or idiosyncratic risk, irrespective whether the latter two measures are estimated

based on the CAPM or Carhart (1997) model. One could still argue that the total risk of

team managed funds, even though being insignificant statistically, is large in economic

sense, reaching almost 1% per year (0.0797*12). However, the two metrics of the

idiosyncratic risk, IdVol1 and IdVol4, that have different signs as well as economically

similar yet small exposures to the market portfolio, Mrk1 and Mrk4 imply at best

potential exposure to non-conventional risk measures. Indeed, we observe that team-

managed funds load more on small firms and high book-to-market firms: the coefficients

on SMB and HML are both positive and significant. Among control variables, the most

consistent results for market risk are that we find that it increases for large funds and

funds with higher turnover rates. Also, we note that fund age has negative and almost

everywhere statistically significant impact on risk across most of its measures except

momentum.

101

In sum, Table 9 illustrates that the impact of group-decision making on fund risk

taking behavior is not very straightforward. Team-managed funds do not seem to take

more market risk, and their volatility, both total and risk-adjusted, is also non-excessive,

but they may expose themselves more to other possible measures of risk than single-

managed funds.

Next, we look if team management is associated with specific fund characteristics

using the regression setting below:

titititititi eFEControlsMgrControlsFundTeamddFundChar ,,3,21,1,10, __ +++++= − δδδ ,

(10)

where FundChari,t is one of fund’s i characteristics at time t. Four fund characteristics are

relevant for our analysis: Expenses, turnover, fund size and net flows. Clearly, in these

regression models, our set of fund-level control variables must depend on the fund

characteristic in question.

Table 10 reports the results of tests based on Eq. (10). The table has 12 columns,

three regression specifications per each fund characteristic. Columns 1-2 show the results

for fund expenses. Consistent with Table 1 data, we find that team-managed funds are

generally cheaper for investors. This result is significant with fund-level controls but,

with the sample reduction after the addition of manager-level controls, drops to

insignificance. Columns 3-4 show the results for fund turnover. We observe that team

management drastically reduces the trading frequency of funds and this drop is

statistically significant. For instance, in economic terms, an average team-managed fund

reduces annual turnover by 12.4% relative to a single-managed fund with similar fund

and manager characteristics. Columns 5-6 show the results for fund size. A priori, one

can think that larger funds are more likely to have teams of portfolio managers. However,

102

just like Table 1 provides no clear signs that multiple-manager funds are usually larger,

the estimation results in the current table that account for control variables also give no

convincing support for any relation between team management and fund size.35 Finally,

in columns 7-8, we show the impact of teams on generating fund flows. In these tests, we

follow Sirri and Tufano (1998) and, besides controlling for the standard set of fund

characteristics, also the lagged unconditional alpha, α(4U)i,t-1, and the lagged flows to

funds with the same investment objective, Obj Flowsi,t-1. We find that team-managed

funds are able to generate significantly higher net flows to their respective funds. This

statistically significant result becomes even stronger after the incorporation of managerial

controls in the last column of the table. Our finding that team-managed funds increase

fund flows is also consistent with recent trend of the increase in the proportion of

multiple-manager funds.

3.7 Conclusions

In this paper, we revisit the question on the benefits of collective versus

individual decision making. Few studies exist in economics literature that estimate the

impact of a team on worker productivity and risk taking in rather indirect ways basing

their findings on relatively limited data. Using detailed managerial-level data from

mutual funds allows one to directly observe any differences in various aspects of

performance and risk preferences between single-managed and team-managed funds.

However, prior research in this area has been largely relying on CRSP dataset and the

prevailing conclusion has been multiple-manager funds perform no better if not worse

than solo-manager ones.

35 Our set of fund-level control variables also include lagged fund size as in Chevalier and Ellison (1999a).

103

We use mutual fund data from Morningstar and first meticulously show that there

exist large discrepancies in managerial structure reporting between this database and

CRSP. This misspecification averages about 20% per year over our sample period of

1992-2010. More importantly, using more reliable Morningstar data we provide

compelling evidence that team management has on average a positive impact on fund

risk-adjusted returns across all fund investment objectives except aggressive growth. In

these tests, we are able to control for a range of fund-level and manager-specific

characteristics.

We further show that the influence of a group decision making on fund

performance is non-linear in team size and is not uniform across all geographic locations.

Funds benefit the most from a team work of three portfolio managers. This may indicate

the potential trade-off between the benefits of collective wisdom and increasing

coordination and/or free-rider issues that become more problematic in larger groups.

Also, the benefits of team management are strongly present among funds in financial

centers but not outside those locations. This outcome is consistent with the idea that

larger cities provide wider opportunities for learning and knowledge spillovers, so the

potential contribution of each manager to fund activities in larger cities is higher than in

smaller towns. We observe that team management practice in financial centers is

effective among funds with more homogeneous managers along education and age

dimensions, possibly reflecting the benefits of more alignment in career concerns.

Finally, we show that among other benefits of team-managed funds are substantially

lower turnover and ability to attract new money flows into their funds.

104

Table 1 Summary statistics of mutual funds management structure

Panel A: Distribution of single and team-managed funds 1 Manager 2 Managers 3 Managers 4 Managers 5+ Managers Number Percent Number Percent Number Percent Number Percent Number Percent 1992 519 67 145 19 70 9 17 2 29 4 1993 584 63 202 22 78 8 20 2 39 4 1994 672 64 243 23 85 8 23 2 35 3 1995 729 61 273 23 115 10 30 3 45 4 1996 767 57 350 26 121 9 57 4 46 4 1997 859 56 399 26 161 11 63 4 48 3 1998 921 53 449 26 210 12 67 4 84 5 1999 961 51 494 26 258 14 81 5 99 6 2000 987 49 587 29 253 12 90 5 116 6 2001 1004 47 602 28 272 13 115 6 134 7 2002 1000 46 647 30 283 13 120 6 137 7 2003 971 44 662 30 287 13 145 7 161 8 2004 876 39 659 30 320 14 174 9 196 10 2005 832 35 698 29 335 14 226 11 300 14 2006 802 33 731 30 352 14 222 11 346 16 2007 776 31 748 30 363 15 247 12 333 16 2008 776 32 732 30 356 15 243 12 327 16 2009 719 31 691 30 392 17 189 9 315 16 2010 622 29 666 31 398 19 164 9 293 16 Total 15377 43 9978 28 4709 13 2293 7 3083 10

Panel B: Fund performance of single and team-managed funds

1 Manager 2 Managers 3 Managers 4 Managers 5+ Managers Mean S.D. Mean S.D. Mean S.D. Mean S.D. Mean S.D. OAR 0.001 1.347 0.015 1.283 0.018 1.157 0.048 1.480 0.037 0.975 Diff 0.014 0.017 0.047 0.036 p-value (0.447) (0.470) (0.147) (0.176) αααα(4U) -0.042 0.796 -0.031 0.765 -0.006 0.738 -0.029 0.788 -0.005 0.603 Diff 0.011 0.036** 0.013 0.037** p-value (0.342) (0.017) (0.528) (0.031) αααα(4C) -0.006 0.857 -0.003 0.822 0.033 0.793 0.009 0.866 0.018 0.659 Diff 0.003 0.039** 0.015 0.024 p-value (0.806) (0.016) (0.498) (0.188)

105

Table 1 (continued) Panel C: Fund characteristics of single and team-managed funds

1 Manager 2 Managers 3 Managers 4 Managers 5+ Managers

Mean S.D. Mean S.D. Mean S.D. Mean S.D. Mean S.D.

Volatility 4.728 2.567 4.820 2.647 4.981 2.638 4.756 2.701 4.715 2.262

TNA 914 3,800 667 2,030 864 2,690 941 3,450 2,310 10,300

Fund Age 10.240 12.569 10.208 12.185 10.201 12.209 9.193 10.514 10.615 11.446

Turnover 0.913 0.843 0.856 0.698 0.906 0.745 0.828 0.630 0.807 0.627

Expenses 1.316 0.475 1.292 0.437 1.270 0.424 1.244 0.410 1.178 0.407

Panel D: Fund manager characteristics of single and team-managed funds

1 Manager 2 Managers 3 Managers 4 Managers 5+ Managers

Mean S.D. Mean S.D. Mean S.D. Mean S.D. Mean S.D.

Tenure 4.42 4.80 3.83 3.60 3.67 3.20 3.52 3.2 3.61 2.9

SAT 1157.44 139.12 1146.17 116.13 1143.16 99.95 1139.86 93.01 1145.23 79.91

MBA 0.53 0.50 0.70 0.46 0.80 0.40 0.87 0.34 0.95 0.23

Mage 45.90 9.56 44.99 8.83 44.34 8.53 44.32 8.53 44.48 7.09

This table gives the summary statistics of domestic equity mutual funds in the United States from 1992 to 2010. Panel A reports the number (and percentage) of funds managed by one, two, three, four, and five (or more) fund managers each year. Panel B report the mean and standard deviation of three fund performance measures. OAR (%, per month) is investment objective adjusted fund return, which is the difference between the average monthly net fund return for fund i in year t and the average monthly fund return of all funds in the matched investment objective in year t. α(4U) and α(4C) are the monthly risk-adjusted net fund returns using unconditional and conditional versions of Carhart (1997) four-factor model, respectively. The panel also shows the difference in performance test results between each group of team-managed funds and single-managed funds. Panel C reports the mean and standard deviation of different fund characteristics over the entire sample period. Volatility (%) is the standard deviation of monthly fund returns over the past 12 months for fund i in year t. TNA ($, millions) is the total net asset under management of fund i in year t. Fund Age (years) is the difference between fund i’s inception year and the current year t. Turnover is the minimum of aggregated sales or aggregated purchases of securities of the year divided by the average 12-month total net assets of the fund. Expenses (%) is the annual total expense ratio of the fund i in year t. Panel D reports fund manager characteristics following Chevalier and Ellison (1999). Tenure (years) is the number of years the fund manager remains with the fund i at time t. SAT is the SAT score of matriculates of the fund manager’s undergraduate institution. MBA is defined as a dummy variable which equals one when a fund manager (or at least one of the team members) has MBA degree and zero otherwise. MAge (years) is the fund manager’s age at current year t. Important note: In case of teams, we simply take the average for each of these characteristics: Tenure, SAT and MAge.

106

Table 2

Misspecification in management structure: CRSP versus Morningstar Misspecification

Year Matched # Funds

CRSP Morningstar Single(CRSP) - Team(MS) Team(CRSP) - Single(MS) # Misspecified Funds

% Matched Sample % Single % Team % Single % Team # Funds % Single(CRSP) # Funds % Team(CRSP)

1992 582 80.76 19.24 67.87 32.13 89 18.94 14 12.50 103 17.70 1993 720 81.94 18.06 64.58 35.42 147 24.92 22 16.92 169 23.47 1994 835 79.64 20.36 63.35 36.65 176 26.47 40 23.53 216 25.87 1995 946 78.22 21.78 61.42 38.58 196 26.49 37 17.96 233 24.63 1996 1040 69.04 30.96 58.17 41.83 173 24.09 60 18.63 233 22.40 1997 1238 63.25 36.75 56.54 43.46 166 21.20 83 18.24 249 20.11 1998 1560 60.90 39.10 54.17 45.83 222 23.37 117 19.18 339 21.73 1999 1668 54.02 45.98 50.84 49.16 177 19.64 124 16.17 301 18.05 2000 1678 52.26 47.74 48.63 51.37 197 22.46 136 16.98 333 19.85 2001 1798 50.17 49.83 47.94 52.06 183 20.29 143 15.96 326 18.13 2002 1864 47.64 52.36 46.51 53.49 190 21.40 169 17.32 359 19.26 2003 1933 42.42 57.58 44.28 55.72 145 17.68 181 16.26 326 16.86 2004 1940 33.04 66.96 40.21 59.79 116 18.10 255 19.63 371 19.12 2005 2015 33.20 66.80 35.33 64.67 184 27.50 227 16.86 411 20.40 2006 2068 33.70 66.30 33.46 66.54 203 29.12 198 14.44 401 19.39 2007 2129 31.38 68.62 31.75 68.25 122 18.26 130 8.90 252 11.84 2008 2110 30.19 69.81 32.65 67.35 122 19.15 174 11.81 296 14.03 2009 1928 30.39 69.61 31.64 68.36 116 19.80 140 10.43 256 13.28 2010 1866 30.98 69.02 29.80 70.20 105 18.17 83 6.44 188 10.08

This table describes the nature and extent of misspecification in the management structure of the U.S. domestic equity mutual funds from 1992 to 2010. Using a matched sample of mutual funds in the CRSP and Morningstar (MS) mutual fund database, the first columns in the table report the percentage of mutual funds classified as reporting one manager name (Single-managed), reporting two or more manager names (Team-managed) in both databases by year. In both cases the unit of observation is the mutual fund, not the fund share class. Columns seven to twelve report the extent of management structure misspecification in the matched sample by year. Column seven reports the number of funds that are classified as single-managed in CRSP but are team-managed in MS in the same calendar year. Column eight reports these misspecified funds as a percentage of all funds classified as single-managed in CRSP. Similarly, column nine reports the number of funds that are identified as team-managed in CRSP but are single-managed in MS. Column ten reports these misspecified funds as a percentage of all funds classified as team-managed in CRSP. Columns eleven and twelve report the total number of misspecified funds and express it as a percentage of total matched sample each year.

107

Table 3 Effect on team management on fund performance: CRSP versus Morningstar

Panel A: Full matched sample analysis CRSP Morningstar αααα(4U) αααα(4U) αααα(4C) αααα(4C) αααα(4U) αααα(4U) αααα(4C) αααα(4C) Team -0.0012 -0.0108 -0.0033 -0.0058 0.0134 0.0247 0.0127 0.0340** (0.912) (0.475) (0.777) (0.728) (0.204) (0.106) (0.266) (0.039) Fund Sizei,t-1 -0.0270*** -0.0316*** -0.0260*** -0.0243*** -0.0272*** -0.0321*** -0.0262*** -0.0248*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Fund Agei,t -0.0035 -0.0166* -0.0092 -0.0304*** -0.0031 -0.0175* -0.0087 -0.0316*** (0.629) (0.076) (0.244) (0.003) (0.672) (0.061) (0.269) (0.002) Family Sizei,t-1 0.0122*** 0.0128*** 0.0125*** 0.0085** 0.0123*** 0.0134*** 0.0126*** 0.0092** (0.000) (0.001) (0.000) (0.049) (0.000) (0.001) (0.000) (0.032) Expensesi,t-1 -0.0573*** -0.0585*** -0.0457*** -0.0472** -0.0568*** -0.0571*** -0.0451*** -0.0455** (0.000) (0.002) (0.005) (0.022) (0.000) (0.003) (0.006) (0.027) Turnoveri,t-1 -0.0271*** -0.0224* -0.0050 0.0137 -0.0268*** -0.0212* -0.0047 0.0153 (0.003) (0.078) (0.622) (0.334) (0.003) (0.097) (0.644) (0.280) Flowsi,t-1 -0.0043 -0.0026 -0.0057* -0.0057 -0.0043 -0.0026 -0.0057* -0.0057 (0.150) (0.468) (0.075) (0.126) (0.153) (0.464) (0.077) (0.123) Performancei,t-1 0.0948*** 0.1027*** 0.0782*** 0.0809*** 0.0948*** 0.1025*** 0.0782*** 0.0805*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Tenurei,t 0.0038** 0.0044** 0.0045*** 0.0052*** (0.022) (0.013) (0.007) (0.003) SATi,t 0.0220*** 0.0196*** 0.0230*** 0.0208*** (0.000) (0.002) (0.000) (0.001) MBAi,t 0.0093 -0.0021 0.0015 -0.0112 (0.587) (0.917) (0.928) (0.571) MAgei,t -0.0838** -0.1010** -0.0789* -0.0952** (0.038) (0.017) (0.051) (0.025) Constant 0.1668** 0.3841** 0.1248 0.4854** 0.1611** 0.3366* 0.1189 0.4286** (0.025) (0.033) (0.132) (0.014) (0.032) (0.065) (0.153) (0.031) Time & Obj. FE Yes Yes Yes Yes Yes Yes Yes Yes Cluster (Fund) Yes Yes Yes Yes Yes Yes Yes Yes R2 (%) 12.65 13.33 12.76 13.31 12.66 13.35 12.76 13.34 Obs. 18,437 10,982 18,437 10,982 18,437 10,982 18,437 10,982 Team (MS-CRSP) = 0 0.0146*** 0.0355*** 0.0160*** 0.0398*** p-value (0.000) (0.000) (0.000) (0.000)

108

Table 3 (continued) Panel B: Sub-period analysis with αααα(4U) as the dependent variable

CRSP Morningstar

1992-1999 2000-2010 1992-1999 2000-2010 Team -0.0020 0.0213 0.0015 -0.0165 0.0215 0.0520* 0.0122 0.0153 (0.941) (0.515) (0.892) (0.314) (0.400) (0.086) (0.298) (0.402) Fund Controls Yes Yes Yes Yes Yes Yes Yes Yes Mgr. Controls No Yes No Yes No Yes No Yes Constant Yes Yes Yes Yes Yes Yes Yes Yes Time & Obj. FE Yes Yes Yes Yes Yes Yes Yes Yes Cluster (Fund) Yes Yes Yes Yes Yes Yes Yes Yes R2 (%) 5.42 5.66 15.13 16.37 5.43 5.74 15.14 16.37 Obs. 3,626 2,618 14,811 8,364 3,626 2,618 14,811 8,364 Team (MS-CRSP) = 0 0.0235*** 0.0307*** 0.0107*** 0.0319*** p-value (0.000) (0.000) (0.000) (0.000) This table compares the effect of management structure on fund performance across CRSP and Morningstar databases using a panel regression approach on matched sample from 1992 to 2010. Panel A reports regression estimates of the matched funds across full sample period using both databases, while Panel B reports regression estimates of the matched funds across two sub-periods. In Panel A, the dependent variable includes two performance measures, α(4U) and α(4C), which are the monthly risk-adjusted net fund returns using unconditional and conditional versions of Carhart (1997) four-factor model, respectively. In Panel B the dependent variable is α(4U). The independent variable of interest is Team, defined as a dummy variable which equals one if the fund has two (or more) fund managers and zero if it has only one fund manager at the end of calendar year. Other independent variables include various fund and manager characteristics as controls. Fund Size is the log of total net assets under management of the fund. Fund Age is the log of the difference between the fund’s inception year and the current year. Family Size is the log of total net asset under management of the fund’s family. Expenses is the annual total expense ratio of the fund. Turnover is the minimum of aggregated sales or aggregated purchases of securities of the year divided by the average 12-month total net assets of the fund. Flows is the net growth in total net assets under management of the fund over the past year. Performance is the corresponding lagged fund performance measure, α(4U) or α(4C). Tenure is the number of years the fund manager remains with the fund. SAT is the SAT score (divided by 100) of matriculates of the fund manager’s undergraduate institution. MBA is defined as a dummy variable which equals one when a fund manager (or at least one of the team members) has MBA degree and zero otherwise. Manager Age is the log of fund manager’s age in current year. All regression specifications include time and investment objective fixed effects (FE), and the standard errors are clustered by fund. Each regression model also reports the p-values of coefficients, the number of observations and the adjusted R2. Team (MS-CRSP) is the hypothesis that slope coefficients on Team in the corresponding Morningstar and CRSP estimations are the same and p-value is the p-value of this test. ***, ** and * indicate significance at the 1%, 5%, and 10% levels, respectively.

109

Table 4 Effect of team management of fund performance

Panel A: Tests with net (expense-adjusted) returns

OAR αααα(4U) αααα(4C)

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Teami,t 0.0237 0.0128 0.0308* 0.0216** 0.0181* 0.0320** 0.0189* 0.0184 0.0381** (0.154) (0.332) (0.094) (0.031) (0.100) (0.043) (0.081) (0.118) (0.025)

Fund Sizei,t-1 -0.0384*** -0.0555*** -0.0211*** -0.0253*** -0.0225*** -0.0211*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)

Fund Agei,t 0.0048 -0.0048 -0.0138* -0.0284*** -0.0142* -0.0382*** (0.631) (0.696) (0.080) (0.004) (0.093) (0.000)

Family Sizei,t-1 0.0126*** 0.0208*** 0.0104*** 0.0110*** 0.0106*** 0.0080* (0.000) (0.000) (0.000) (0.005) (0.001) (0.060)

Expensesi,t-1 -0.0214 -0.0462* -0.0414** -0.0532*** -0.0369** -0.0472** (0.311) (0.056) (0.011) (0.007) (0.041) (0.028)

Turnoveri,t-1 0.0377*** 0.0372** -0.0279*** -0.0241* -0.0196* -0.0046 (0.005) (0.031) (0.004) (0.072) (0.069) (0.746)

Volatilityi,t-1 -0.0307** -0.0063 -0.0112** -0.0068 0.0235*** 0.0261*** (0.017) (0.786) (0.033) (0.339) (0.000) (0.000)

Flowsi,t-1 -0.0025 -0.0057 -0.0001 0.0008 -0.0027 -0.0033 (0.411) (0.169) (0.974) (0.818) (0.346) (0.356)

FCi 0.0110 -0.0007 -0.0047 -0.0079 -0.0051 0.0104 (0.378) (0.965) (0.663) (0.571) (0.663) (0.491)

Tenurei,t 0.0078*** 0.0060*** 0.0071*** (0.001) (0.001) (0.000)

SATi,t 0.0321*** 0.0212*** 0.0188***

(0.000) (0.001) (0.005)

MBAi,t 0.0293 0.0005 -0.0091 (0.174) (0.979) (0.645)

MAgei,t -0.0017 -0.1032** -0.1310*** (0.971) (0.012) (0.002)

Constant 0.0098 0.5517*** 0.2223 -0.0556** 0.1569** 0.4361** -0.0996*** 0.0213 0.4687** (0.751) (0.000) (0.296) (0.049) (0.044) (0.019) (0.001) (0.801) (0.019)

Time & Obj. FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Cluster (Fund) Yes Yes Yes Yes Yes Yes Yes Yes Yes R2 (%) -0.04 1.93 2.99 11.02 11.90 12.77 11.09 12.31 13.25

Obs. 31,440 20,565 12,135 26,703 19,781 11,646 26,703 19,781 11,646

110

Panel B: Tests with gross (expense-unadjusted) returns

This table shows the effect of management structure on fund performance using the Morningstar U.S. domestic equity mutual fund sample from 1992 to 2010. It reports the estimates from panel regressions of fund performance on management structure (team versus single) and other controls. Panel A shows test results with net (expense-adjusted) returns; Panel B – with gross (expense-unadjusted) returns. The dependent variable includes three performance measures: OAR, α(4U), and α(4C). OAR is the difference between the average monthly net fund return for the fund in year t and the average monthly net fund returns of all funds in the matched investment objective in year t. α(4U) and α(4C) are the monthly risk-adjusted net fund returns using unconditional and conditional versions of Carhart (1997) four-factor model, respectively. The independent variable of interest is Team, defined as a dummy variable which equals one if the fund has two (or more) fund managers and zero if the fund has only one fund manager at the end of calendar year. Other independent variables include various fund and manager characteristics as controls. Fund Size is the log of total net assets under management of the fund. Fund Age is the log of the difference between the fund’s inception year and the current year. Family Size is the log of total net asset under management of the fund’s family. Expenses is the annual total expense ratio of the fund. Turnover is the minimum of aggregated sales or aggregated purchases of securities of the year divided by the average 12-month total net assets of the fund. Flows is the net growth in total net assets under management of the fund over the past year. Volatility (%) is the standard deviation of monthly net fund returns over the past 12 months for the fund. FC is the dummy variable which equals one if the fund is in a financial center and zero otherwise. Financial center funds have headquarters located within 50 miles of Boston, Chicago, Los Angeles, New York, Philadelphia, or San Francisco. Tenure is the number of years the fund manager remains with the fund. SAT is the SAT score (divided by 100) of matriculates of the fund manager’s undergraduate institution. MBA is defined as a dummy variable which equals one when a fund manager (or at least one of the team members) has MBA degree and zero otherwise. MAge is the log of fund manager’s age in current year. All regression specifications include time and investment objective fixed effects and the standard errors are clustered by fund. Each regression model also reports the p-values of coefficients, the number of observations and the adjusted R2. ***, ** and * indicate significance at the 1%, 5%, and 10% levels, respectively.

OAR αααα(4U) αααα(4C)

(1) (2) (3) (4) (5) (6) (7) (8) (9) Teami,t 0.0179 0.0091 0.0288 0.0172* 0.0159 0.0302* 0.0145 0.0163 0.0362** (0.273) (0.482) (0.112) (0.082) (0.142) (0.056) (0.177) (0.161) (0.032) Fund Controls No Yes Yes No Yes Yes No Yes Yes Mgr. Controls No No Yes No No Yes No No Yes Constant Yes Yes Yes Yes Yes Yes Yes Yes Yes Time & Obj. FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Cluster (Fund) Yes Yes Yes Yes Yes Yes Yes Yes Yes R2 (%) -0.05 2.16 3.22 10.86 11.91 12.82 10.97 12.54 13.49 Obs. 31,440 20,565 12,135 26,703 19,781 11,646 26,703 19,781 11,646

111

Table 5 Effect of team management of fund performance by investment objective

Aggressive Growth Growth Growth & Income Equity Income

αααα(4U) αααα(4C) αααα(4U) αααα(4C) αααα(4U) αααα(4C) αααα(4U) αααα(4C) Teami,t -0.0010 -0.0179 0.0305 0.0388* 0.0736*** 0.0833*** 0.0765* 0.0804* (0.981) (0.696) (0.151) (0.082) (0.005) (0.003) (0.050) (0.051) Fund Controls Yes Yes Yes Yes Yes Yes Yes Yes Mgr. Controls Yes Yes Yes Yes Yes Yes Yes Yes Constant Yes Yes Yes Yes Yes Yes Yes Yes Time & Obj. FE Yes Yes Yes Yes Yes Yes Yes Yes Cluster (Fund) Yes Yes Yes Yes Yes Yes Yes Yes R2 (%) 15.23 15.06 13.84 13.42 15.66 18.00 16.04 18.42 Obs. 2,402 2,402 6,908 6,908 1,761 1,761 575 575 This table shows the effect of management structure on fund performance using the Morningstar U.S. domestic equity mutual fund sample from 1992 to 2010. It reports the estimates from panel regressions of fund performance on management structure (team versus single) and other controls and other controls across four different MS investment objective categories: Aggressive Growth, Growth, Growth & Income, and Equity Income. The dependent variable includes two performance measures, α(4U), and α(4C), which are the monthly risk-adjusted net fund returns using unconditional and conditional versions of Carhart (1997) four-factor model, respectively. The independent variable of interest is Team, defined as a dummy variable which equals one if the fund has two (or more) fund managers and zero if the fund has only one fund manager at the end of calendar year. Other independent variables include various fund and manager characteristics as controls and are the same as in Table 4. All regression specifications include time and investment objective fixed effects and the standard errors are clustered by fund. Each regression model also reports the p-values of coefficients, the number of observations and the adjusted R2. ***, ** and * indicate significance at the 1%, 5%, and 10% levels, respectively.

112

Table 6 Effect of team size on fund performance

αααα(4U) αααα(4C)

(1) (2) (3) (4) (5) (6)

2 Managers 0.0121 0.0124 0.0307* 0.0058 0.0073 0.0304 (0.317) (0.336) (0.091) (0.653) (0.600) (0.119) 3 Managers 0.0359** 0.0320** 0.0405* 0.0384** 0.0388** 0.0499** (0.015) (0.045) (0.065) (0.016) (0.021) (0.032) 4 Managers 0.0155 -0.0068 0.0154 0.0230 0.0052 0.0392 (0.516) (0.737) (0.526) (0.373) (0.802) (0.126) 5+ Managers 0.0305** 0.0328* 0.0291 0.0236 0.0307* 0.0361 (0.043) (0.050) (0.184) (0.151) (0.093) (0.123) Fund Controls No Yes Yes No Yes Yes Mgr. Controls No No Yes No No Yes

Constant Yes Yes Yes Yes Yes Yes

Time & Obj. FE Yes Yes Yes Yes Yes Yes Cluster (Fund) Yes Yes Yes Yes Yes Yes R2 (%) 11.13 11.96 12.83 11.20 12.39 13.28 Obs. 25,908 19,555 11,534 25,908 19,555 11,534 This table shows the effect of team size on fund performance using the Morningstar U.S. domestic equity mutual fund sample from 1992 to 2010. It reports the estimates from panel regressions of fund performance on team size and other controls. The dependent variable includes two risk-adjusted performance measures, α(4U) and α(4C). α(4U) and α(4C) are the monthly risk-adjusted net fund returns using unconditional and conditional versions of Carhart (1997) four-factor model, respectively. 2 Managers is a dummy variable which equals one if the fund has two fund managers at the end of calendar year and zero otherwise; 3 Managers is a dummy variable which equals one if the fund has three fund managers at the end of calendar year and zero otherwise; 4 Managers is a dummy variable which equals one if the fund has four fund managers at the end of calendar year and zero otherwise; 5+ Managers is a dummy variable which equals one if the fund has five (or more) fund managers at the end of calendar year and zero otherwise. Other independent variables include various fund and manager characteristics as controls and are the same as in Table 4. All regression specifications include time and investment objective fixed effects and the standard errors are clustered by fund. Each regression model also reports the p-values of coefficients, the number of observations and the adjusted R2. ***, ** and * indicate significance at the 1%, 5%, and 10% levels, respectively.

113

Table 7 Interaction of team and location on fund performance

αααα(4U) αααα(4C) (1) (2) (3) (4) (5) (6) Teami,t 0.0095 -0.0052 0.0144 -0.0089 -0.0152 0.0092 (0.572) (0.774) (0.551) (0.612) (0.417) (0.725) Teami,t ×FCi 0.0222 0.0400* 0.0290 0.0469** 0.0578** 0.0475 (0.286) (0.076) (0.326) (0.033) (0.016) (0.138) FCi -0.0124 -0.0290 -0.0278 -0.0267 -0.0401** -0.0223 (0.459) (0.127) (0.284) (0.129) (0.046) (0.431) Fund Controls No Yes Yes No Yes Yes Mgr. Controls No No Yes No No Yes

Constant Yes Yes Yes Yes Yes Yes

Time & Obj. FE Yes Yes Yes Yes Yes Yes Cluster (Fund) Yes Yes Yes Yes Yes Yes R2 (%) 11.29 11.91 12.77 11.33 12.34 13.26 Obs. 24,714 19,781 11,646 24,714 19,781 11,646 F-test: FC (Team - Single) 0.0317** 0.0348** 0.0434** 0.0380*** 0.0426*** 0.0567*** p-value (0.013) (0.011) (0.026) (0.007) (0.004) (0.007) F-test: NFC(Team - Single) 0.0095 -0.0052 0.0144 -0.0089 -0.0152 0.0092 p-value (0.572) (0.774) (0.551) (0.612) (0.417) (0.725) This table shows the impact of management structure and fund location interaction has on fund performance using the Morningstar U.S. domestic equity mutual fund sample from 1992 to 2010. It reports the estimates from panel regressions of fund performance on Team and Financial Center location and other controls. The dependent variable includes two performance measures, α(4U) and α(4C), which are the monthly risk-adjusted net fund returns using unconditional and conditional versions of Carhart (1997) four-factor model, respectively. Independent variables of interest are Team×FC, Team, and FC, where Team is defined as a dummy variable which equals one if the fund has two (or more) fund managers and zero otherwise, while FC is a dummy variable which equals one if the fund is located in a financial center and zero otherwise. Financial center funds have their advisors located within 50 miles of Boston, Chicago, Los Angeles, New York, Philadelphia, or San Francisco. Other independent variables are defined as in Table 4. All regression specifications include time and investment objective fixed effects and the standard errors are clustered by fund. Each regression model also reports the p-values of coefficients, the number of observations and the adjusted R2. ***, ** and * indicate significance at the 1%, 5%, and 10% levels, respectively.

114

Table 8 Effect of team diversity on fund performance across geographic locations

Financial Centers Non-Financial Centers αααα(4U) αααα(4C) αααα(4U) αααα(4C)

(1) (2) (3) (4) (5) (6) (7) (8) Tenure Diversityi,t -0.0290 0.0093 -0.0657 -0.0229 0.0351 0.0772 0.0805 0.1225** (0.515) (0.838) (0.167) (0.643) (0.499) (0.166) (0.155) (0.049) SAT Diversityi,t -0.6243** -0.6434** -0.8084*** -1.0662*** 0.0325 0.0066 -0.0126 -0.0395 (0.032) (0.019) (0.007) (0.000) (0.938) (0.988) (0.977) (0.933) MAge Diversityi,t -0.4225** -0.3695** -0.5429*** -0.4976*** 0.2100 0.2642 0.1042 0.1605 (0.027) (0.047) (0.002) (0.003) (0.202) (0.110) (0.552) (0.353) Team Sizei,t 0.0047 0.0193 -0.0488** -0.0235 (0.815) (0.377) (0.037) (0.357) Fund Controls No Yes No Yes No Yes No Yes Mgr. Controls No Yes No Yes No Yes No Yes Constant Yes Yes Yes Yes Yes Yes Yes Yes Time & Obj. FE Yes Yes Yes Yes Yes Yes Yes Yes Cluster (Fund) Yes Yes Yes Yes Yes Yes Yes Yes R2 (%) 15.14% 16.59% 17.40% 18.35% 13.18% 15.79% 15.39% 17.83% Obs. 1,924 1,667 1,924 1,667 1,350 1,214 1,350 1,214

This table shows the impact of team diversity on fund performance across fund locations using the Morningstar U.S. domestic equity mutual fund sample from 1992 to 2010. It reports the estimates from panel regressions of fund performance on three team diversity measures across funds located in financial centers and other places. The dependent variable includes two performance measures, α(4U) and α(4C) which are the monthly risk-adjusted net fund returns using unconditional and conditional versions of Carhart (1997) four-factor model, respectively. Independent variables of interest are Tenure Diversity, measured by the coefficient of variation of all managers’ tenure with the fund in a team; SAT Diversity, measured by the coefficient of variation of all managers’ SAT scores within a team; and Manager Age (MAge) Diversity, measured by coefficient of variation of all fund managers’ age (in years) within a team. Team Size equals the number of fund managers within a team in a given year. For teams with four or more managers the Team Size equals four. Other independent variables are defined as in Table 4. Financial center funds have their advisors located within 50 miles of Boston, Chicago, Los Angeles, New York, Philadelphia, or San Francisco. All regression specifications include time and investment objective fixed effects and the standard errors are clustered by fund. Each regression model also reports the p-values of coefficients, the number of observations and the adjusted R2. ***, ** and * indicate significance at the 1%, 5%, and 10% levels, respectively.

115

Table 9 Effect of team management on risk-taking behavior

CAPM Unconditional Carhart Model Total Risk Mrk1 IdoVol1 Mrk4 SMB HML MOM IdVol4

Teami,t 0.0797 0.0102 0.0569 0.0096 0.0267* 0.0301** -0.0030 -0.0061 (0.180) (0.385) (0.137) (0.216) (0.056) (0.044) (0.699) (0.787) Fund Sizei,t-1 0.0490*** 0.0144*** -0.0146 0.0087*** -0.0037 -0.0064 0.0008 -0.0200*** (0.002) (0.000) (0.201) (0.000) (0.296) (0.124) (0.735) (0.005) Fund Agei,t -0.1058*** -0.0128* -0.0901*** -0.0026 -0.0252*** -0.0226*** 0.0103** -0.0428*** (0.002) (0.062) (0.000) (0.580) (0.001) (0.007) (0.027) (0.001) Expensesi,t-1 0.3180*** 0.0509*** 0.3614*** 0.0042 0.1106*** -0.0330* -0.0010 0.2253*** (0.000) (0.000) (0.000) (0.637) (0.000) (0.071) (0.919) (0.000) Turnoveri,t-1 0.3300*** 0.0806*** 0.2109*** 0.0330*** 0.0798*** -0.0888*** 0.0668*** 0.1001*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) FCi, 0.0013 0.0105 -0.0292 0.0080 0.0059 -0.0066 0.0029 -0.0299 (0.980) (0.338) (0.422) (0.267) (0.666) (0.662) (0.707) (0.163) SATi,t -0.0159 -0.0011 -0.0355** 0.0018 -0.0049 0.0066 -0.0049 -0.0146 (0.478) (0.817) (0.023) (0.558) (0.389) (0.322) (0.108) (0.132) Tenurei,t 0.0032 -0.0014 0.0225*** -0.0027*** 0.0064*** 0.0039** -0.0015 0.0144*** (0.608) (0.297) (0.000) (0.003) (0.000) (0.022) (0.109) (0.000) MBAi,t -0.1129 -0.0194 -0.0977** -0.0003 -0.0171 0.0110 0.0143 -0.0859*** (0.140) (0.187) (0.046) (0.973) (0.283) (0.525) (0.122) (0.004) MAgei,t -0.1954 -0.0528* -0.0724 -0.0213 -0.0680** 0.0705* -0.0279 -0.0142 (0.175) (0.075) (0.448) (0.276) (0.044) (0.065) (0.159) (0.801) Constant Yes Yes Yes Yes Yes Yes Yes Yes Time & Obj. FE

Yes Yes Yes Yes Yes Yes Yes Yes

Cluster (Fund) Yes Yes Yes Yes Yes Yes Yes Yes R2 (%) 58.00 18.56 45.46 6.90 31.71 12.12 10.03 33.15 Obs. 12,891 12,286 12,286 12,286 12,286 12,286 12,286 12,286 This table shows the effect of management structure on risk-taking behavior of mutual funds using the Morningstar (MS) U.S. domestic equity mutual fund sample from 1992 to 2010. The table reports the estimates from panel regressions of fund risk-taking on Team and other controls. The dependent variable includes different measures of risks. Total Risk is defined as the standard deviation of monthly net fund returns over the past twelve months. Mrk1 is the market risk defined as the coefficient of the market portfolio based on the CAPM performance evaluation model. IdVol1 is the standard deviation of the fund’s residual return from the CAPM model. Mrk4, SMB, HML, and UMD are coefficients of market, size, book-to-market, and momentum portfolios based on the Carhart (1997) four-factor performance evaluation model. IdVol4 is the standard deviation of the fund’s residual return from the Carhart (1997) model. The independent variable of interest is Team, defined as a dummy variable which equals one if the fund has two (or more) fund managers and zero if the fund has only one fund manager at the end of calendar year. Other independent variables include various fund and manager characteristics as controls. Fund Size is the log of total net assets under management of the fund. Fund Age is the log of the difference between the fund’s inception year and the current year. Family Size is the log of total net asset under management of the fund’s family. Expenses is the annual total expense ratio of the fund. Turnover is the minimum of aggregated sales or aggregated purchases of securities of the year divided by the average 12-month total net assets of the fund. FC is the dummy variable which equals one if the fund is in a financial center and zero otherwise. Financial center funds have headquarters located within 50 miles of Boston, Chicago, Los Angeles, New York, Philadelphia, or San Francisco. Tenure is the number of years the fund manager remains with the fund. SAT is the SAT score (divided by 100) of matriculates of the fund manager’s undergraduate institution. MBA is defined as a dummy variable which equals one when a fund manager (or at least one of the team members) has MBA degree and zero otherwise. MAge is the log of fund manager’s age in current year. All regression specifications include time and investment objective fixed effects and the standard errors are clustered by fund. Each regression model also reports the p-values of coefficients, the number of observations and the adjusted R2. ***, ** and * indicate significance at the 1%, 5%, and 10% levels, respectively.

116

Table 10 Effect of team management on fund characteristics

Expenses Turnover Fund Size Flows

(1) (2) (3) (4) (5) (6) (7) (8) Teami,t -0.0252** -0.0198 -0.0551** -0.1243*** 0.0071 0.0216* 0.0449** 0.0757** (0.043) (0.273) (0.012) (0.000) (0.410) (0.081) (0.043) (0.020) Fund Sizei,t-1 -0.0474*** -0.0527*** -5.7641*** -5.0287*** 0.9300*** 0.9313*** -0.2245*** -0.2210*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Fund Agei,t 0.0064 -0.0117 -1.4391 1.7573 -0.0341*** -0.0456*** -0.0733*** -0.0993*** (0.544) (0.397) (0.350) (0.310) (0.000) (0.000) (0.000) (0.000) Family Sizei,t-1 -0.0245*** -0.0134*** 3.6593*** 1.7497** 0.0305*** 0.0294*** 0.0922*** 0.0843*** (0.000) (0.007) (0.000) (0.021) (0.000) (0.000) (0.000) (0.000) Turnoveri,t-1 0.0004*** 0.0003** -0.0167** -0.0046 -0.0139 0.0109 (0.000) (0.013) (0.012) (0.572) (0.417) (0.656) Volatilityi,t-1 0.0244*** 0.0231*** 6.3197*** 6.5111*** -0.0130*** -0.0156*** -0.0055 0.0007 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.493) (0.936) Flowsi,t-1 -0.0049*** -0.0052*** -0.0378 0.1958 0.0367*** 0.0413*** (0.001) (0.010) (0.885) (0.557) (0.000) (0.000) FCi 0.0154 0.0060 12.2349*** 12.8597*** -0.0118 -0.0047 -0.0142 -0.0028 (0.320) (0.753) (0.000) (0.000) (0.194) (0.685) (0.547) (0.923) Expensesi,t-1 13.8426*** 10.5228*** -0.0543*** -0.0601*** -0.1940*** -0.1882*** (0.000) (0.001) (0.000) (0.000) (0.000) (0.000) αααα(4U)i,t-1 0.1982*** 0.2158*** (0.000) (0.000) Obj. Flowi,t-1 0.1732** 0.1598 (0.024) (0.118) Mgr. controls Yes Yes Yes Yes Constant Yes Yes Yes Yes Yes Yes Yes Yes Time & Obj. FE Yes Yes Yes Yes Yes Yes Yes Yes Cluster (Fund) Yes Yes Yes Yes Yes Yes Yes Yes R2 (%) 17.96 16.57 9.38 13.57 93.18 92.96 9.59 9.87 Obs. 22,407 13,279 20,854 12,312 20,566 12,136 20,565 12,135

117

Table 10 (continued)

This table shows the effect of management structure on different fund characteristics of U.S. domestic equity mutual funds from 1992 to 2010. The table reports panel regressions estimates of different fund characteristics on Team and other controls. The dependent variable includes: Expenses, defined as the annual total expense ratio of the fund; Turnover, defined as the minimum of aggregated sales or aggregated purchases of securities of the year divided by the average 12-month total net assets of the fund; Fund Size, defined as the log of total net assets under management of the fund; and Flows, defined as the net growth in total net assets under management of the fund over the past year. The independent variable of interest is Team, defined as a dummy variable which equals one if the fund has two (or more) fund managers and zero if the fund has only one fund manager at the end of calendar year. Other independent variables include various fund and manager characteristics as controls. Fund Age is the log of the difference between the fund’s inception year and the current year. α(4U) is the monthly risk-adjusted net fund return using Carhart (1997) four-factor model. Family Size is the log of total net asset under management of the fund’s family. Volatility (%) is the standard deviation of monthly net fund returns over the past 12 months for the fund. FC is the dummy variable which equals one if the fund is a financial center fund and zero otherwise. Financial center funds have headquarters located within 50 miles of Boston, Chicago, Los Angeles, New York, Philadelphia, or San Francisco. Tenure is the number of years the fund manager remains with the fund. SAT is the SAT score of matriculates of the fund manager’s undergraduate institution. MBA is defined as a dummy variable which equals one when a fund manager (or at least one of the team members) has MBA degree and zero otherwise. MAge is the log of fund manager’s age in current year. All regression specifications include time and investment objective fixed effects and the standard errors are clustered by fund. Each regression model also reports the p-values of coefficients, the number of observations and the adjusted R2. ***, ** and * indicate significance at the 1%, 5%, and 10% levels, respectively.

118

Figure 1. Evolution of mutual fund management structure from 1992 to 2010.

This figure shows the percentage of single-managed and team-managed funds along with the total number of funds in our sample for 1992 to 2010. The left-hand side vertical axis represents the percentage of single- and team-managed funds out of the total funds in our sample each year. The right-hand side vertical axis represents the total of funds in our sample each year. The horizontal axis represents each year included in our sample.

0

10

20

30

40

50

60

70

80

1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

Perc

ent

of

Funds

0

500

1000

1500

2000

2500

3000

Num

ber

of

Funds

Single-managed funds Team-managed funds Total number of funds

119

Chapter 4

Deception and Managerial Structure:

A Joint Study of Portfolio Pumping and

Window Dressing Practices

120

4.1 Introduction

In this article, we study whether organizational structure impacts the likelihood of

deception.36 In particular, we examine whether team-based organizations deter agents

from engaging in deceptive and unethical behavior. We achieve this goal by examining

the extent of portfolio pumping and window dressing – the two fund practices that are

considered illegal or quasi-illegal – among single-managed and team-managed funds in

the U.S. mutual fund industry (see Appendix for the U.S. Securities and Exchange

Commission’s (SEC) litigation cases involving both these practices). Portfolio pumping

is a practice when fund managers artificially inflate their year-end (and often quarter-

end) performance by placing large orders on existing holdings (see Zweig, 1997; Carhart,

Kaniel, Musto, and Reed 2002; Bernhardt and Davies, 2005). Market regulators regard

this action as illegal. Another type of fund managers’ trading behavior that is largely

perceived as being dishonest, although without a formal status of illegal activity, is

window dressing. It is a practice when fund managers buy (sell) stocks that recently

performed well (bad) just before the fund’s holdings are made public to give the

impression that they’ve been holding good stocks in their portfolios for a while (e.g., see

Lakonishok, Shleifer, Thaler, Vishny, 1991; Sias and Starks, 1997; He, Ng, and Wang,

2004; Ng and Wang, 2004; Meier and Schaumburg, 2006; Agarwal, Gay, and Ling,

2011).

The relation between organizational structure and agents’ incentives is well

known (see Arrow, 1974). The predisposition of team-based organizations to cheat may

be lower than that of individuals due to three main factors: social, economic, and

psychological.37 First, teams may increase the cost of cheating by greater social pressure

36 There are several types of deception (lies). We follow Gneezy’s (2005) classification and assume for the purposes of our work that deception is an increase of the payoff to the liar and decrease of the payoff to the liar’s counterpart. 37 We use words deception and cheating interchangeably. In economics literature, cheating is often associated with the free-rider problem (shirking) rather than illegal behavior.

121

and peers monitoring (e.g., see Arnott and Stiglitz, 1991; Mas and Moretti, 2009).38

Teams may also reduce the benefits from cheating since they divide their total production

output among all members, thus transforming high-powered incentives into low-powered

ones and reducing each individual member’s monetary incentives to cheat (e.g., see Ma,

Moore, and Turnbull, 1988; Kandel and Lazear, 1992; Acemoglu, Kremer, and Mian,

2008). Finally, when working in teams, individuals may experience higher moral

pressures such as guilt aversion (see Charness and Dufwenberg, 2006). However, almost

all the aforementioned studies are theoretical, and few existing empirical results are

related only to free-riding. There is no any empirical evidence on the relation between

organizational structure and deception.

Our data comes from Morningstar Direct and covers actively managed U.S.

domestic equity mutual funds from January 2, 1992 to December 31, 2010. These funds

belong to one of the four investment objectives: aggressive growth, growth, growth &

income, and equity income. To properly disentangle any potential impact of managerial

structure on fund deception tactics, we account for several fund-specific variables, such

as investment objective, total assets under management, trading frequency, and

performance – all of which may have incremental impact on the propensity to cheat.

Using standard methodologies in the literature, we test portfolio pumping on daily

returns and window dressing on quarterly returns to match the frequency in reporting of

fund portfolio holdings.

First, we examine the impact of managerial structure on portfolio pumping.

Consistent with prior findings, we find strong evidence of portfolio pumping over the

whole sample period. However, it is present more among single-managed than team-

38 In 2010, the SEC started a whistleblower program to encourage people to report fraud within their own organization by providing them with financial incentives and protection (see http://www.sec.gov/whistleblower). This program makes fraud reporting easier, thus deterring illegal trading activities.

122

managed funds, at both the year-end and quarter-ends. The average end-of-year

(beginning-of-year) daily excess returns among single-managed funds are different from

those of the rest of the year by 32bps (-26bps). The same returns for team-managed funds

differ from those of single-managed ones by -0.11bps and 0.04bps for the end-of-year

and beginning-of-year, respectively. These differences are highly statistically significant.

Moreover, we find that the evidence of portfolio pumping is decreasing with the number

of fund managers in a team. For example, two-manager funds show -8bps and 2bps

differences with single-managed funds for the end-of-year and beginning-of-year daily

returns, respectively, but these differences for funds with five or more managers are -

16bps and 11bps, respectively. This constitutes 50% lower returns on the last day of the

year and 42% higher return in the first day of the year, respectively, among funds with

five or more managers as compared to single-managed funds. The decreasing relation

between the extent of portfolio pumping and team size holds across all fund investment

objectives, but it is more profound among aggressive growth and growth funds.

We analyze how team size affects portfolio pumping across different fund sizes.

We find evidence of portfolio pumping across all fund sizes, but its extent is again

smaller among team-managed funds in all instances, at both the year- as well as quarter-

ends, with the exception of the year-end for the smallest quartile of funds. However, even

in this size quartile, funds experience much weaker pumping when managed by teams of

five or more managers. In a similar vein, portfolio pumping activity is present across all

turnover quartiles, but, as expected, it is stronger among high turnover funds. Importantly

though, even among funds with the lowest turnover, team-management consisting of five

or more people helps reduce this activity.

We also relate the extent of portfolio pumping across managerial structures for

funds with different prior performances. In contrast to previous studies (e.g., Carhart et

al., 2002), we find that portfolio pumping can be at least as profound among the worst

123

performing funds as among the best performers. Notably, we observe that the strongest

evidence of portfolio pumping occurs among the worst-performing single-managed

funds. It appears that managers of these funds have the highest incentives for making the

performance of their funds look better and very low pressures for not undertaking any

illegal trading activities.

Next, we examine window dressing. Overall, team-managed funds sell less

extreme losing stocks and buy less extreme winning stocks than their single-managed

counterparts. When looked at the quarterly trading activity, we find, consistent with prior

studies, that the evidence of window dressing, more specific trading activity at the year-

end, is concentrated in sells of extreme losing stocks. Importantly, while single-managed

funds are involved in more selling of bad performing stocks in the fourth quarter of the

year compared to the previous three quarters, the trading activity of these types of stocks

among team-managed ones is similar throughout the year. Moreover, the difference in

selling intensity of the most underperforming stocks between team-managed and single-

managed funds is decreasing (increasing in magnitude) in team size across all quarters. In

addition, we document an absence of window dressing practice among team-managed

funds across all four fund investment objectives. Yet, the selling intensity of the worst

performing stocks among single-managed funds is always higher in at the end-of-the-

year.

Then, given the fact that some of the characteristics of team-managed funds may

again affect their ability to window dress, we analyze the impact of fund size, turnover,

and performance on the established negative relation between the propensity to window

dress and team size. We observe that the selling intensity of the worst performing

securities even among the largest single-managed funds is prone to window dressing.

The difference in their selling intensity of bad stocks is again economically and

statistically larger in the fourth quarter of the year as compared to the first three quarters.

124

The largest team-managed funds have similar selling intensity throughout the year.

Similarly, when we look at fund turnover quartiles, we continue observing the same

pattern with selling of the worst performance stocks: economically sizable window

dressing is observed among single-managed funds but not team-managed. Across fund

performance quartiles we find that the selling rate of poorly performing stocks in team-

managed funds is significantly lower in the fourth quarter compared to the first three

quarters of the year in the middle performance range. Yet, in the same performance

brackets, single-managed funds significantly increase their trading towards the end of

year.

Finally, we conduct our window dressing tests over the period of 1996-2000.

Greenwood and Nagel (2009) mention that window dressing is likely to be observed

more during the Dot-com bubble since many funds at that time would be interested in

reporting that they were holding high-performing telecom and internet stocks. Indeed,

our results show that the extent of window dressing is more prevalent in the late 1990s,

yet, again only among single-managed funds.

Thus, our study shows that team management significantly inhibits managers’

drive to deceive. In particular, deceptive tactics of investment managers such as

portfolio pumping and window dressing are largely the prerogatives of single-manager

funds. The three potential explanations for our results can be linked to peer monitoring,

monetary incentives, and psychological factors. Due to data limitations, we are unable to

differentiate across these explanations for more ethical behavior of team-managed funds.

Nonetheless, our findings provide a clear picture that, whatever the reason may be, team-

management in the mutual fund industry is beneficial to both individual investors and the

public as a whole, at least, from to the perspective of reduction in illegal or quasi-illegal

activities.

125

The rest of the article is organized as follows. Section 2 provides motivation for

differences in incentives to deceive between individuals and teams. Section 3 examines

the extent of portfolio pumping among team-managed and single-managed funds.

Section 4 focuses on the relation between managerial structure and window dressing. The

descriptive statistics of data are present within each of these two sections. Section 5

concludes.

4.2 Motivation and Related Literature

It has been understood for a long time that groups behave differently from

individuals. Le Bon (1896) is the first to introduce the idea of a “group mind”,

differentiating it from a mind of a single individual. Since then, many studies in various

fields, including economics and finance, have attempted to compare group and individual

decision making in terms of performance and risk taking. The related literature is divided

into two camps. On the one side, many papers argue that decisions made within groups

are inferior to those made by individuals.39 This may result from extreme decisions by a

dominant team member or a reduction in critical thinking in each team member for the

sake of more conformity across the entire group. In economics, the negative effects of

groups are usually associated with cheating (free-riding) by some group members that

lead to the loss of productivity (e.g., see Holmstrom, 1982; Rasmusen, 1987; Nalbantian

and Schotter, 1997). Likewise, few finance studies find no benefits of teamwork in

enhancing the performance of professional money managers (e.g., Chen, Hong, Huang,

and Kubik, 2004; Massa, Reuter, and Zitzewitz, 2010).

On the other side, some theoretical and empirical evidence shows that teams

outperform individuals and do not take excessive risks. For example, Sah and Stiglitz

39 See, for example, Moscovici and Zavalloni (1969) and Janis (1982).

126

(1986, 1991) and Sharpe (1981) show that the opinion of a team is the average of

opinions of each team member, and, therefore, teams help diversifying individuals’

opinions. Barry and Starks (1984) provide a theoretical setting suggesting that teams in

investment funds may reduce portfolio risk. However, there are very few empirical

studies that find evidence of increased performance in teams based on theories of opinion

and risk diversification in groups.40 Hamilton, Nickerson, and Owan (2003) find that

teams increase productivity, and that this increase is more apparent among earliest team

members. Patel and Sarkissian (2012) use equity mutual fund data and observe that team-

managed funds outperform single-managed ones without resorting to extra risk taking.41

An additional dimension where a team behavior may differ from that of an

individual is the likelihood of deceptive and illegal behavior. The probability of

involving in an illegal action depends on its costs and benefits. Becker (1968) argues that

criminals are rational agents: they weight the cost and benefit of a crime before actually

committing it. Different organizational structures imply different costs and benefits of

deceptive behavior. First, teams increase social pressures and provide an easy ground for

mutual supervision, thus reducing deception opportunities and increasing the cost of

deviating from the “right” behavior. Arnott and Stiglitz (1991) argue that peers

monitoring is important in labor markets because workers (e.g., portfolio managers) are

often in a better position to monitor their co-workers (e.g., co-portfolio managers) than

are employers (e.g., a fund company or investors). Mas and Moretti (2009) conclude that

workers in a team experience disutility if they are observed behaving selfishly by their

peers, irrespective whether their co-workers’ sanctions are formal or informal. Second,

teams provide a different compensation structure, thus reducing the benefits of cheating.

Ma et al. (1988) argue that the principal can overcome the problem of “cheating” by

40 The majority of existing studies in this area are experimental (e.g., see Bornstein and Yaniv, 1998; Bone et al., 1999; Barber et al., 2003; Cooper and Kagel, 2004; and Blinder and Morgan, 2005; among others). 41 Other papers, such as Adams and Ferreira (2010) and Bar, Kempf, and Ruenzi (2011), find that teams take less extreme decisions than individuals, but they provide no evidence of better performance in teams.

127

linking one agent’s contract to other agent’s output. Kandel and Lazear (1992) find that

peer pressure and monitoring are more effective when profits are shared in by all

members of the organization (e.g., partnership). Acemoglu et al. (2008), using a career

concerns model and some evidence from pension funds, show that when agents are

provided with high-powered incentives, which induce them to apply sufficient amount of

good efforts, these incentives may also encourage agents to use bad efforts to improve

their observed performance. Acemoglu et al. (2008) further argue that working in teams

may transform high-powered incentives of individual team members into low-powered

ones resulting in lesser frequency of bad behavior.42 Third, there are psychological

factors that explain why individuals may be less inclined to deceive when working in a

team. For instance, Charness and Dufwenberg (2006), using a model based on contract

theory, find that when agents exhibit guilt aversion then communication among them

may enhance their trustworthy behavior. Thus, teams are less likely to deceive than

individuals. Moreover, the positive effects of peers monitoring and social pressures, as

well as economic costs of deceptive behavior are likely to increase with team size.

Therefore, the larger is the team the less likely that its members will cheat.

Various experimental studies generally support a view that groups (not crowds)

behave smarter than individuals.43 Studies like Bornstein and Yaniv (1998), Cooper and

Kagel (2005), and Sutter (2009) argue that groups are more rational than individuals. In

particular, groups behave more strategically and learn more quickly to act strategically

than individuals. Charness et al. (2007) find that the more salient are the links among

group members, the more often group members select actions producing the largest

payments for both themselves and the rest of the group.

42 Jacob and Levitt (2003) show that high-powered incentives often lead to such distortions in behavior as cheating. 43 Large groups with loosely defined ties may show off the negative characteristics of a crowd.

128

The next problem is to find a data source which would allow us to confidently

test whether there is a relation between organizational structure and deception. In this

respect, mutual funds provide one of the best datasets for analyzing the impact of group

decision making on the likelihood and extent of deceptive actions. There are several

reasons for that. First, at present, mutual funds provide one of the largest single sources

of occupational data. They represent the largest cross-sectional sample of firms within a

single industry and provide the longest time-series of data covering about two decades of

observations with sufficiently large details. Second, the proportion of team-managed and

single-managed funds is such that it allows one to make inferences about their relative

performance, risk-taking, and deception propensity while dealing with comparable

samples. Third, fund industry in general, and mutual funds in particular, have been

identified with certain types of practices that are perceived illegal or quasi-illegal on the

part of investors and government bodies. These are two trading actions: portfolio

pumping and window dressing.44 Portfolio pumping is a practice when fund managers

place large orders on existing holdings to artificially inflate their year-end and often

quarter-end performance. This is a very distractive practice for investors since after a

temporary gain in performance, stock prices usually fall back to previous levels once the

impact from the positive price pressure is over. Zweig (1997), Carhart et al. (2002), and

Bernhardt and Davies (2005) document that portfolio pumping is quite common across

various fund types. Market regulators regard this action as illegal and the SEC has

changed few portfolio managers with this behavior. (See the Appendix, sub-section A.1,

for some recent SEC cases related to portfolio pumping.)

44 Another type of investment activity that may have ill-motivated trades is risk-shifting (e.g., see Brown, Harlow, and Starks, 1996; Chevalier and Ellison, 1997; Goetzmann, Ingersoll, Spiegel, and Welch, 2007; Huang, Sialm, and Zhang, 2011). It is a practice of changing the fund’s risk exposure to maximize its performance by the reporting date and to attract additional fund flows. However, this activity is not regarded as unethical, since even good performing funds with skilled managers may resort to risk-shifting for simply maximizing the returns on their investments. Also, to the best of our knowledge, there has not been any official litigation case related to risk-shifting (unlike portfolio pumping and window dressing). Therefore, we do not consider this trading tactic as universally unethical and do not examine it in our study.

129

Another type of fund managers’ trading behavior that perceived as being

dishonest is window dressing. It is a practice when fund managers buy (sell) stocks that

recently performed well (bad) just before the fund’s holdings are made public giving the

impression that they’ve been holding good stocks in their portfolios for a while.

Lakonishok et al. (1991) observed these trading patterns among pension funds. He et al.

(2004) as well as Ng and Wang (2004) document similar behavior across a variety of

financial institutions. Sias and Starks (1997) find the evidence of the turn-of-the-year

effects not only among institutional investors but also individuals. Yet, they attribute

their findings not so much to window dressing as to tax-loss-selling. It is indeed more

difficult to identify whether funds have been involved in window dressing activities than

whether they practice portfolio pumping. However, in spite of this difficulty, there are

cases when the SEC has explicitly cited window dressing tactics while charging portfolio

managers with illegal trading schemes. (See the Appendix, sub-section A.2, for the

SEC’s case related to window dressing.) Moreover, with recent increase in fund

performance reporting at the quarterly frequency, the incidences of window dressing are

likely to occur not only at the end of the year.

4.3 Portfolio Pumping and Managerial Structure

4.3.1 The Detail of Portfolio Pumping Phenomenon and its Estimation Methodology

It has been widely documented in both academic literature and professional

reports that returns of stocks and equity funds exhibit various seasonality effects. Of

particular interest was the observation of unusually large fund returns at the New Year’s

Eve. While some evidence of artificial stock price inflation by fund managers has been

making the headlines in popular press in the 1990s (see Zweig, 1997), Carhart et al.

130

(2002) offered the first comprehensive study on both yearly and quarterly fund

performance manipulation. This frequency of abnormal patterns in fund returns coincides

with the reporting frequencies among mutual funds.45 Ippolito (1992), Sirri and Tufano

(1998) and others observe that money flows into the best performing funds. Since it is in

fund managers’ compensation interests to have increasing inflows to their funds, they

have strong incentive to inflate their performance by the reporting dates. Expectedly,

Meier and Schaumburg (2006) and Agarwal et al. (2011) show that funds with poor

recent performance are more likely to engage in window dressing.

There also exist some theoretical studies that justify the existence of portfolio

pumping and/or provide some predictions on the extent of this phenomenon across funds.

Bhattacharyya and Nanda (2009) develop an equilibrium model where managers have

incentives to alter the closing prices of their security holdings. Bernhardt and Davies

(2009) show that portfolio pumping is persistent, that is, those mutual funds who are

involved in portfolio pumping in one quarter are likely to do it again in the following

quarter.

Since the results of Carhart et al. (2002) became public around the year 2000, the

SEC began scrutinizing suspicious fund trading activities and enforcing the existing

trading laws much better. As a result of these actions, in the June of 2001, SEC filed the

first fraud charges against a fund manager for market manipulation and portfolio

pumping. Duong and Meschke (2011) find a substantial decrease in portfolio pumping

activity afterwards. However, our focus is not so much on the changes in magnitude of

speculative price manipulation over time but on the cross-sectional differences in the

extent of portfolio pumping related trading between team-managed and single-managed

funds.

45 Reflecting the position that more transparency is better than less, in May 2004 the SEC increased the required portfolio disclosure frequency from semi-annual to quarterly frequency. Prior to 2004 funds could voluntarily report on a quarterly basis.

131

The primary source of mutual fund data is the Morningstar Direct database.46 Our

sample covers actively managed U.S. domestic open-end equity mutual funds from

January 2, 1992 to December 31, 2010. We focus on funds that belong to aggressive

growth (includes small company), growth, growth & income, and equity income

investment objectives. We exclude all sector, balanced, international, and index funds

from our analysis. The dataset includes daily fund returns (net of expenses), funds size,

measured by the total net assets (TNA) under management of the fund at the end of

calendar year, and fund turnover, computed as the minimum of aggregated sales or

purchases of securities in a year divided by the average 12-month TNAs of the fund. All

fund characteristics except fund returns are at individual fund level, so we aggregate

mutual fund share class level return observations to individual fund level using a unique

fund identifier in Morningstar Direct. To obtain fund returns in excess of a benchmark,

we subtract the daily fund returns from the returns of S&P 500 index. To minimize the

effect of outliers on our analysis, we winsorize daily excess fund returns at 1% and 99%

levels.

The Morningstar database also contains names of fund managers responsible for

day-to-day management of fund each year and the exact joining and leaving dates of fund

managers. We determine the managerial structure of funds based on the total number of

fund managers at the end of the calendar year. If a fund names only one fund manager at

the end of calendar year, we classify that fund as single-managed for that year. If a fund

names two or more fund managers, we classify that fund as team-managed. Further, we

divide team-managed funds into funds with two, three, four, and five (or move) distinct

fund managers at the end of calendar year, denoted 2FM, 3FM, 4FM, and 5+FM,

respectively. We remove all fund-years where fund manager names or tenure dates are

missing. Our final sample covers 3,252 unique funds with 7,053,857 daily observations.

46 Several recent studies show that Morningstar data on managerial structure is superior to that of CRSP (see Massa et al., 2010; Patel and Sarkissian, 2012).

132

To estimate the managerial structure impact on portfolio pumping, we amend the

Carhart et al. (2002) methodology. Specifically, our regression model is:

,,13121110

98765

43210,

tiittt

tttitit

ititititti

eTeambMBEGbMENDbQBEGb

QENDbYBEGbYENDbTeamMBEGbTeamMENDb

TeamQBEGbTeamQENDbTeamYBEGbTeamYENDbbr

++++

++++×+×

+×+×+×+×+=

(1)

where ri,t is the fund i daily return (net of expenses) in excess of the daily S&P500 index

return. Independent variables include Teami, which is a dummy variable equals to one if

fund i has two (or more) fund managers and zero otherwise. YEND and YBEG are the last

and first trading day of year dummies, respectively. QEND is the last trading day of the

quarter, that is, March, June or September dummy; QBEG is the first trading day of the

quarter, that is, April, July or October dummy. MEND is the last trading day of February,

April, May, July, August, October or November dummy; and MBEG is the first trading

day of February, March, May, June, August, September, November or December

dummy. The coefficients of primary interest are those on the interaction terms of Team

and YEND, b1, YBEG, b2, QEND, b3, and QBEG, b4. They indicate how much fund

returns around the end of the year and end of quarters are different from average returns

during the rest of the year.

Table 1 shows the summary statistics of data that we use to measure the level of

portfolio pumping. It reports the mean and standard deviation of daily excess returns of

funds across various team sizes, as well as their respective number of observations

(divided by 1000). It also gives these return statistics across funds with different

investment objectives. The average returns of team-managed funds are higher than

single-managed ones, consistent with evidence in Patel and Sarkissian (2012). Across

investment objectives, the highest returns are observed, as expected, in the aggressive

growth category while the lowest – in growth & income and equity income categories.

The average excess returns of team-managed funds are higher than single-managed ones

133

for all four fund investment objectives. The table also provides summary statistics on two

important fund characteristics that are known to be related to fund returns. These are

fund size, measured by the total net assets under management of the fund at the end of

calendar year. We can see that the average fund size is somewhat larger for team-

managed funds. However, the average fund size for funds managed by less than five

people is, in fact, smaller than that of single-managed fund. Only funds with five or more

managers have substantially larger total net assets than their single-manager counterparts.

The turnover of team-managed funds is lower than that of single-managed ones for any

team size.

4.3.2 Test Results

Table 2 shows the aggregate results on the portfolio pumping activity across

funds with different managerial structures. It reports the end-of-year and beginning-of-

year, end-of-quarter, and beginning-of-quarter coefficients and their corresponding p-

values (in parentheses) for single-managed funds, across all team-managed funds, and

separately for funds with various team sizes. The standard errors are clustered by fund.

Consistent with Carhart et al. (2002) and others, we find strong evidence of portfolio

pumping both around the year-end and quarter-ends. Importantly, the most profound

evidence of this seasonal trading activity is concentrated in single-managed funds. The

average daily excess returns of single-managed funds at the year-end and quarter-ends

differ from their returns during the rest of the year by 32bps and 20bps, respectively. The

same return differences among team-managed funds are lower by 11bps and 5bps. This

implies, that team-managed funds earn about 33% (25%) lower additional returns on the

last day of the year (quarter) compared to single-managed funds. Similar pattern is

detected for the beginning-of-year and beginning-of-quarter returns. These returns for

single-managed funds differ from those of the rest of the year by -26bps and -21bps,

134

respectively, but are less negative by about 4bps for team-managed funds. Furthermore,

we notice that the evidence of portfolio pumping is decreasing with the number of fund

managers in a team. For example, two-manager funds exhibit -8bps and 2bps differences

with single-managed funds for the end-of-year and beginning-of-year daily returns,

respectively. Yet, the same differences for funds with five managers are substantially

larger standing at -16bps and 11bps, respectively. This implies that funds with five or

more managers experience 50% lower returns on the last day of the year and 42% higher

returns in the first day of the year that funds with only one manager.

The next issue is to identify whether portfolio pumping activity patterns

documented in Table 3 are present across all fund investment categories, various fund

sizes and occur irrespective of average fund turnover. Table 3 presents results on

portfolio pumping evidence across four investment objectives and various fund manager

team sizes. We notice that the largest extent of pumping occurs, not surprisingly, among

aggressive growth funds, followed by growth funds. For example, the daily excess

returns at the year-end and year-beginning among single-managed aggressive growth

funds are 50bps and -49bps, respectively. The quarter-end and quarter-beginning results

are similar, standing at 40bps and -41bps. The magnitude of end-of-year (beginning-of-

year) returns among team-managed aggressive growth funds is less positive (negative)

than that of single-managed ones by 15bps (7bps). Similar to the overall picture in Table

2, we again observe that the dampening effect of team management on pumping activity

increases with team size. Among aggressive growth funds, the strength of portfolio

pumping for funds with five of more managers is lower by about 35% than that for

single-managed funds: the daily returns for funds with five and more managers around

December 31 – January 1 are -22bps and 17bps (compare with 50bps and -49bps for

single-managed funds, respectively).

135

Table 4 reports the estimation results of daily fund returns around the year- and

quarter-ends for four fund size quartiles. Quartile 1 includes the smallest 25% of all

funds, while Quartile 4 – the largest 25%. The evidence of portfolio pumping is present

across all fund sizes, but, consistent with previous literature, the spread in the magnitudes

between positive returns in the last day of the year and negative returns in the first day of

the year are larger for the smallest size quartile. Yet, this pattern is observed only among

single-managed funds. Team-managed funds, even the smallest ones, exhibit much less

positive year-end returns and much less negative beginning-of-year ones. Similar

patterns are observed for returns around the quarter-ends. As before, the effect of team-

management on the reduction of portfolio pumping is increasing with team size across all

fund size quartiles. The takeaway from this table is that not only the smallest but also the

largest funds are not immune from portfolio pumping, but that team management

alleviates this problem significantly, especially when teams are comprised of more than

three people.

Table 5 shows the estimation results of daily fund returns around the year- and

quarter-ends for four fund turnover quartiles. Quartile 1 includes funds with the lowest

25% turnover, while Quartile 4 – funds with the highest 25% turnover. Again, not

unexpectedly, we find more evidence of portfolio pumping among funds with the highest

turnover rates based on returns around both year- and quarter-ends. That is, that are

accustomed to trade a lot, are also more prone to portfolio pumping activity. The table

shows that the management team impact on returns around the year-ends and quarter-

ends are similar to all previous tables: portfolio pumping significantly decreases with

team size.

Thus, Tables 2-5 show that the propensity of funds managers to artificially alter

returns at the end of reporting periods, including the year-end is substantially reduced

when managers act within a team, especially teams of four, five or more managers.

136

Importantly, this reduction is not a characteristic of any one fund investment objective,

size or turnover. What remains to be seen is how group decision making impacts the

extent of portfolio pumping across funds with various performance. We accomplish this

in Figures 1 and 2.47

Figure 1 shows fund returns around the year-end depending on fund performance

for all funds and separately for single-managed and team-managed funds. The upper part

of the plot shows daily excess fund returns on the last trading day of the year; the lower

part – on the first trading day of the year. Fund performance is measured from the first

trading day of the year to the second-to-last day of the same year and is split into 20

performance bins by 5% each. Similar to Carhart et al. (2002), we find a U-shaped

pattern between the end-of-year returns and fund performance for the whole sample of

funds, as well as for the sub-samples of single- managed and team-managed funds. We

also observe an inverse U-shaped pattern between the beginning-of-year returns and fund

performance. What is different from Carhart et al. (2002) is that, unlike them, we find

that the evidence of portfolio pumping is more profound not among high-performing

funds, but those in the lowest yearly performance. For example, for the full sample of

funds, while excess positive returns at the year-end are about the same for funds with

both the worst and the best performance (around 33bps), the year-beginning returns are

markedly lower among the worst performing funds (close to an average of 30bps for the

bottom 15bps of performance) than their best performing counterparts (about -25bps).

Nevertheless, substantial year-end return manipulation evidence among high performing

funds indicates, as other studies also concluded, that funds that have very high likelihood

of being classified as top performers have sufficient incentives to increase that

possibility.

47 We also conduct cross-sectional tests similar to those in Carhart et al. (2002) to analyze whether the relation between fund’s subsequent returns over any two trading days is more negative on the first day of the year and quarter. Our results, consistent with the earlier evidence, show more reverse relation for these days. They are available on request.

137

More interesting for our analysis are the differences in portfolio pumping

between single-managed and team-managed funds. As we can see, for any fund

performance, the year-end returns are higher for single-managed funds than team-

managed ones. The same picture (with minus sign) holds almost throughout all

performance bins for the year-beginning returns: these returns are less negative for team-

managed funds than single-managed ones in all but the second-to-the-best performance

bin. In addition, unlike previous studies, using the fund performance dimension, we can

state that the largest extent of portfolio pumping occurs among the worst-performing

single-managed funds. One can easily explain this. Managers of single-managed funds

anticipating that their funds will fall in the lowest performance percentiles have, on the

one side, very high incentives to make their returns look better and, on the other,

relatively low pressures for not getting involved in any form of unethical or illegal

trading behavior. For these types of funds, the cost of cheating (probability of being

caught) versus the benefit from it (improved fund performance, increased inflows) is

much lower than among analogous team-managed funds. When single-managed funds

are successful in deceiving the public, they can enjoy all the benefits from their unethical

actions themselves, while the benefits of cheating in team-managed funds are shared by

all team members. In addition, while the chances of being caught with illegal trading

activity, ceteris paribus, must be equal across all funds, the cost of cheating in team-

managed funds may still be higher than in single-managed ones due, for instance, to

various psychological factors (see Charness and Dufwenberg, 2006).

Figure 2 provides more refined information on managerial structure relative to

Figure 1. It depicts fund returns around the year-end depending on fund performance

across manager teams of different sizes: single manager, two or three managers, and four,

five, or more managers. For the ease of clarity, the daily excess return data for funds with

two or three managers as well as four, five or more managers are averaged. The curve for

138

single-managed funds is the same as in Figure 1 and is shown for convenience.

Consistent with all our previous tests, we see that the magnitude of portfolio pumping

again diminishes with team size. For instance, the scale of positive returns at the end-of-

year and the beginning-of-year among best performing funds with four, five and more

managers is the lowest among all manager team sizes and is not much different from

similar returns of funds with average performance. Thus, Figures 1 and 2 show that such

illegal trading as activity portfolio pumping is practiced primarily among single-managed

funds, especially those with the worst performance.

4.4 Window Dressing and Managerial Structure

4.4.1 The Details of Window Dressing Phenomenon and its Estimation Methodology

For a long time there is strong consensus among academics, financial industry,

and regulatory bodies that the disclosure of trading activity by professional money

managers is very useful, since, in this way, people are able to exercise more control over

those who handle their investments. As a result, fund managers are required to make their

holdings public on a regular basis. While the idea of providing more disclosure on

holdings to current and potential investors seems noble, it may lead to unintended

consequences on the part of fund managers such as short-term portfolio reshuffling with

the goal of better performing assets to be shown to the public. Although not without

some reservations, many academic studies demonstrate that excessive trading activity of

equity mutual funds around the reporting times such as the end of calendar year as well

as end of quarters can often be classified as window dressing (see Lakonishok et al.,

1991; He et al., 2004; Ng and Wang, 2004). Window dressing is primarily associated

139

with funds selling their badly performed stocks and buying those with recent strong price

appreciation just before presenting their holdings to their clients and shareholders.

We analyze window dressing activity across funds with different managerial

structure using three databases: Morningstar Direct, Thomson-Reuters Mutual Fund

Holdings database and CRSP Stock Price database. The Morningstar Direct database

contains information related to managerial characteristics of mutual funds such as names

of fund managers, their joining and termination dates as well as other important fund

characteristics such as performance, assets under management and portfolio turnover

rate. We focus on U.S. domestic equity mutual funds that belong to aggressive growth,

growth, growth & income, and equity income investment objectives from 1992 to 2010.

We exclude all funds that belong to sector, balanced, international and index fund

categories from our analysis.

The Thomson-Reuters Mutual Fund Holdings database contains, for each fund,

complete stock holdings at the end of a given quarter. The dataset includes name of stock

holdings, number of shares held in the current portfolio for each stock holding and net

changes in shares held since previously reported portfolio. Following Lakonishok et al.

(1991), we focus our analysis on all common stocks traded on the NYSE, AMEX and

NASDAQ. Finally, we obtain stock prices, returns, and other stock related information

for all stocks in our analysis from CRSP.

We link the Morningstar mutual fund sample to the Thomson-Reuters holding

database using MFLINKS. To do this, we first match each fund in the Morningstar

sample to CRSP mutual fund database using individual fund tickers and date of

inception. In cases where the fund ticker information is missing, we use fund names

along with their date of inception for matching purposes. Then using each fund’s unique

identifier in CRSP (CRSP_FUNDNO), we obtain each fund’s unique portfolio identifier

140

(FUNDNO) using MFLINKS file. This matching technique results in 2,238 funds

(almost a 70% match with the sample used in Section 3 on portfolio pumping) with a

total of 332,983 observations.

To determine whether managers engage in window dressing, we follow

Lakonishok et al. (1991). Each quarter we classify stocks held by mutual funds into

performance quintiles based on stock returns over the past year up to the end of that

quarter. The lowest (highest) quintile represents the stocks that have performed the worst

(best) over the past year. Then we estimate the total dollar value of holdings, purchases

and sales for each fund in each performance quintile by multiplying the number of shares

each fund holds, buys, and sells each quarter by the average of beginning and end-of-

quarter stock prices.48 Formally, we define the selling intensity as

( ) ( )( )( ) ( )( )

,,1,,,

,1,,,

∑ −∑

−=

ii kqiHOLDkqiSELL

kqiHOLDkqiSELLIntensitySelling (2)

where SELL(i,q,k) is the dollar value of sales by fund k in quarter q and performance

group i, and HOLD(i,q-1,k) is the dollar value of holdings at the end of the quarter q-1 of

the exact same stocks as those in performance group i in quarter q. SELL and HOLD are

defined using the average of beginning and end of quarter q prices. Similarly, the buying

intensity is defined as

( ) ( )( )( ) ( )( )

,,.,.

,,,,

∑

∑=

i

i

qiHOLDUNIVqiHOLDUNIV

kqiBUYkqiBUYIntensityBuying (3)

where BUY(i,q,k) is the dollar purchases by fund k in performance group i in quarter q

and UNIV.HOLD(i,q) is the value of CRSP universe holding in quarter q in performance

48 It is important to point that we only observe net changes in shares held by a fund over a given quarter and not all trades that it makes during the same quarter.

141

group i. Again, BUY and UNIV.HOLD are computed using the average of beginning and

end of quarter q prices.

Table 6 shows the summary statistics of trading intensity across mutual funds

with different managerial structures. It reports the mean and standard deviation of selling

and buying intensities of funds across various team sizes, as well as their respective

number of observations (divided by 1000). The table also provides summary statistics on

fund excess returns, size, and turnover – three important fund characteristics that may

affect the extent of window dressing practice. Fund excess return is the quarterly excess

fund return (in percent) computed as the difference between the quarterly net fund return

and the quarterly S&P 500 index return. Fund size and turnover are defined as in Table 1

and show similar trends with manager team size as those in that table.

The selling and buying intensities are shown for fund portfolio holdings classified

as Extreme Losers and Extreme Winners. Extreme Losers is defined as the lowest stock

performance quintile each quarter, where stocks have the lowest returns over the past

year up to the end of that quarter. Extreme Winners is defined as the highest stock

performance quintile each quarter, where stocks have the highest returns over the past

year up to the end of that quarter. We can see that single-managed funds have higher

selling intensity for losing stocks, 1.30 versus 1.27. Moreover, the selling intensity of the

worst performing stocks decreases uniformly with team size. A funds with two managers

on average sells 29% more of extreme losers (relative to its holdings of those stocks)

than it sells of all stocks, while a fund with five of more managers sells only 21% more

of those stocks. This constitutes a 30% reduction in the selling intensity of the worst

performing stocks among funds with five or more managers relative to single-managed

funds.

142

Note that the selling intensity of the best performing stocks is relatively stable

across all team sizes, and its magnitude is higher than that of the worst performing

stocks. This implies that funds in general, irrespective of their managerial structure, are

keener on getting rid of recent winning stocks than recent losers. Next, we look at the

buying intensity. Here, we can again observe that fund purchases of extremely bad

performing stocks are similar between single- and team-managed funds, are do not differ

much with team size. Finally, almost the same picture emerges for the practice of fund

buying of extreme losers. Yes, the only difference here is that funds with five or more

portfolio managers report much smaller intensity rate than not only single-managed

funds but also funds with fewer than five managers.

4.4.2 Test Results

Table 7 shows the effect of managerial structure on window dressing. It reports

the equal-weighted average selling and buying intensities of the worst performing and

best performing stocks over funds and years comparing across the first three quarters and

the last quarter of the year. The estimates are for the whole sample, as well as for single-

managed and team-managed funds. The standard errors are shown below the mean

estimates. We also report two difference tests. Diff(Q4-13) is the difference between

quarter 4 and quarters 1-3 across selling and buying intensities, respectively. Diff(Team-

Single) is the difference of means between single and team-managed funds within

quarters 1-3 and quarter 4. The p-values for these tests are in parentheses.

The test results of particular interest are those corresponding to the selling

intensity of extreme losers followed by the buying intensity of extreme winners. We

observe that across all funds the average selling intensity of the worst performing stocks

is similar around the year: the difference between selling of these stocks in the fourth

143

quarter is economically and statistically indistinguishable from that in the first three

quarters of the year. More importantly, the same result holds for team-managed funds –

they show even somewhat lower selling activity in bad stocks at the year-end. However,

single-managed funds behave differently. Their quarter 4 selling rate of the worst

performing securities is higher by 3.1% than that in the first three quarters. That is, in

quarter 4, a single-managed fund on average sells 3.1% more of its extremely bad

performing stocks than in quarters 1-3, relative to fund’s sells of all stocks. This

difference, Diff(Q4-13), is statistically significant almost at the 5% level, similar to

results in Lakonishok et al. (1991) for the overall sample of pension funds.49

It is important to recognize that, as a reflection of average turnover differences

(see Table 6), the overall selling activity of extreme losing stocks (and some buying

activity as well) may be naturally lower among team-managed funds than their single-

managed counterparts. Yet, another notable result of Table 7 is that not only the

difference in selling intensity of the worst performers is significantly lower among team-

managed funds, but the difference in this rate between the two managerial structures is

larger in the fourth quarter of the year. Therefore, we believe these trading differences

across managerial structures are not driven by such considerations as tax-loss selling at

the end of the year: rather they indicate different propensities to window dress between

single-managed and team-managed funds. On the buying side, among the best

performing stocks, we observe that both single-managed and team-managed funds show

significantly less purchasing activity in the fourth quarter of the year as compared with

the first three quarters. Finally, with respect to the results in the top-right and bottom-left

quadrants of the table, we only note that they are not providing any information on the

49 It is important to note is that while calculating the difference between quarter 4 and quarters 1-3, we correct for the repeated observations within the same fund-quarter-return quintile category. Without this correction, our t-statistics become comparable to those in He et al. (2004).

144

existence of window dressing in mutual funds. Besides, all the corresponding difference

tests are insignificant.

In Table 8, we compare the selling and buying intensities of the worst and best

performing stocks across funds with different team sizes. Similar to Table 7, Panel A,

reports the average trading intensities for the two stock performance groups across the

three quarters and the fourth quarter of the year. Panel B shows the difference in means

between funds with two, three, four, and five (or more) fund managers and single-

manager funds within quarters 1-3 and quarter 4. We find that not only team-managed

funds do not sell bad stocks more intensely at the year-end on average, but also that this

evidence is uniform across team-managed funds irrespective of the team size. Only funds

with four managers exhibit somewhat higher selling rate of bad performing stocks in the

fourth quarter compared to the first three ones, but this difference is insignificant and, in

addition, its magnitude is still lower than that of single-managed funds. On top of that

notice from Table 6 that the sub-sample of four-manager funds is the smallest in our

analysis. More interestingly, we observe from Panel B that team-managed funds of any

team size show substantially lower selling intensity than single-managed funds

throughout the year. Also, the reduction in selling rate vis-à-vis single-managed funds is

increasing in team size and is statistically significant among funds with five or more

managers. Finally, as in Table 7, on the buying side, we do not observe any evidence of

specific trading patterns across quarters of the year or the size of fund management.

As with portfolio pumping, the next issue is to see whether window dressing

patterns shown in Tables 6-7 are different across fund investment categories, sizes or

turnover. Table 9 repeats the estimation of selling (Panel A) and buying (Panel B)

intensities for the best and worst performing stocks across four investment objectives and

different managerial structure. Note that due to the sizable sample reduction in

comparison to the overall results in Table 6, we expect fewer occurrences of statistical

145

significance in difference tests even for the same level of coefficients. Across all

investment objectives in Panel A, we find a consistently positive difference in selling

intensity between the fourth quarter and the first three quarters of the year among single-

managed funds. Moreover, while this difference is small among aggressive growth funds

(0.5%), it is larger than 3.5% across the other three fund investment objectives, and, for

growth funds, due to their larger sample size, is also marginally significant. Yet, the

same difference is consistently negative among team-managed funds. Another interesting

observation is that even though we already know that team-managed funds on average

trade less than single-managed ones, and that their selling intensity of bad performing

stocks is lower at the year-end than in the first three quarter-ends, this pattern persist

across all funds investment objectives. The buying intensity results for the best

performing stocks in Panel B again confirm, consistent with our earlier findings and

those of other studies, that there are no evidence of more trading activity of this type at

the end of the year.

Table 10 shows the average selling and buying intensities of single- and team-

managed funds across quarters 1-3 and quarter 4 controlling for fund size (Panel A) and

fund turnover (Panel B). At the end of each quarter, funds are sorted into size quartiles

based on the total net assets under their management. Also at the end of each quarter

funds are sorted into turnover quartiles based on their annual turnover rates. The top

(bottom) quartile is classified as large (small) funds, while the top (bottom) quartile is

classified as high (low) turnover funds. In this table, we report the results for only these

two quartiles of fund sizes and turnover rates. All other measures are defined similarly to

previous tables. In Panel A, we notice an increase in selling intensity of bad performing

stocks among single-managed funds but not team-managed ones at the year-end relative

to the first three quarters in the subsets of both large and small funds. Moreover, this

increase is statistically significant and larger in magnitude for the sub-sample of large

146

funds versus small ones (5% versus 3%). As before, there is no evidence of purchases of

winning stocks at the end of the year across all fund types. In Panel B we see that, on

average, selling intensity is higher among low turnover funds than high turnover ones.

This subset of funds also demonstrates increased selling rates in the fourth quarter of the

year. The difference in the selling intensity of extreme losers for single-managed funds in

the fourth quarter versus the first three quarters is around 3%, although it is insignificant

due to the approximately four-fold reduction in the sample size. Contrary to this

outcome, low turnover team-managed funds at the year-end demonstrate a decrease even

of higher magnitude in the selling intensity of badly performed holdings.

Finally, we want to understand the impact of managerial structure on window

dressing depending on fund performance. Clearly, if a fund shows good performance

prior to reporting dates, its managers do not need to prove to their clients that their

portfolio holdings included good stocks. Also, if fund returns are much below the

industry’s median, then, most likely, investors will be very puzzled by seeing many “hot”

stocks in a fund’s portfolio given its underperformance. Thus, a priori, we expect to see

more window dressing evidence in the middle fund performance groups, where the

benefits of this trading activity are the greatest.

Table 11 shows the results of our tests. It again reports the average selling and

buying intensities of single- and team-managed funds across quarters 1-3 and quarter 4.

The Winner (Loser) funds are those within the top 25% (bottom 25%) of median

performance across all funds in a given year. Fund performance is based on the year to

date net excess fund returns up to and including the last day of the given quarter. That is,

fund performance in quarter 1 is measured based on the net excess fund returns from

January to March; quarter 2 – from January to June; quarter 3 – from January to

September; and quarter 4 for the full calendar year. The rest of the format of Table 11 is

similar to previous tables.

147

First, we find that the selling intensity of loser stocks is higher for single-

managed funds than team-managed ones across all four quartiles of fund performance

and all quarters of the year except in quarter 4 for the best performing funds. Second, we

find significantly higher selling rate of loser stocks in quarter 4 relative to quarters 1-3

only among single-managed funds and only in the middle two quarters of fund

performance. These differences for quartiles 2 and 3 are 5.4% and 7.6%, respectively.

Importantly, for the same performance quartiles, team-managed funds decrease their

selling intensity in quarter 4, and this decrease is even statistically significant for quartile

2 (-4.7% drop with a p-value of 7.4%). Third, on the two extremes of fund performance,

we find no significant changes in the selling intensity of loser stocks at the year-end

relative to quarters 1-3, although the overall change is positive for Winner funds and

negative for Loser funds. These patterns are consistent with the notion that window

dressing is more likely to occur among those funds that can gain the most from this

action, but even in those instances, team-managed funds behave diametrically different

from their single-managed counterparts. As in previous tables, we observe no evidence of

window dressing on the purchasing side.

Overall, Tables 7-11 provide substantial support regarding the impact of

managerial structure on such misleading trading activity of professional money managers

as window dressing. There is one caveat though. Like other authors, we do not assume

that any increase in fund trading activities at the quarter- or year-end is necessarily a

manifestation of window dressing. For example, in Table 11, we observe that winner

funds, both single- and especially team-managed, increase their selling rate of loser

stocks in quarter 4 relative to quarters 1-3. Some of this activity may also be related to

momentum trading or other legitimate strategies. However, cross-sectional differences in

trading activity between single- and team-managed funds that we observe under various

scenarios are likely to reveal the fundamental link between managerial structure and

148

trading practice. This relation is further reinforced by the consistency of our window

dressing results with the earlier ones on portfolio pumping.

4.4.3 The Dot-com Bubble: A Special Case of Window Dressing

Greenwood and Nagel (2009) mention that the window dressing practice is more

likely to occur at the times when there are more pressures on portfolio managers to report

that their holdings include highly publicized stocks or group of stocks. For example,

during the Dot-com bubble at the end of 1990s many funds would be interested in

reporting that they were holding high-performing telecom and internet stocks. We test

this in Table 12, which repeats our estimations from Tables 7 and 8 on the impact of

various managerial structures on window dressing but only over the years of 1996-2000.

Panel A depicts the results across all funds, as well as single-managed and team-managed

funds, while Panel B shows the difference tests between funds with different team sizes

and those with single manager. We observe that the evidence of window dressing is

indeed stronger in the late 1990s, but, importantly, again only among single-managed

funds. First, the selling intensity of extreme losers in the fourth quarter of the year for

single-managed funds is by 6.2% higher than that in the first three quarter of the year.

Note that in spite of the decrease in the sample size, the increase in selling of the worst

performing stocks at the year-end is again statistically significant almost at the 5% level.

As before, the difference in quarterly selling intensity between team-managed and single-

managed funds increases both economically and statistically with team size, especially in

the fourth quarter, reaching the rate of 23% for the difference between funds with five or

more managers and their single-managed counterparts.

Table 12 also shows some evidence of window dressing on the purchasing side,

but again only among single-managed funds: their intensity of buying the best

149

performing stocks at the year-end in comparison to previous quarter trading rate is higher

by almost 3% (see Panel A), which is comparable in magnitude to the selling rate of

worst performers among single-managed funds across the full sample period in Table 7.

The difference tests in Panel B indicate that, similar to selling intensity differences, the

difference in buying intensity between team-managed and single-managed funds also

increases substantially with team size, reached for funds with five managers statistically

significant levels even in this reduced sample.

4.5 Conclusions

In this paper, we use U.S. domestic equity mutual fund and examine the extent of

two trading practices, portfolio pumping and window dressing across funds with

different managerial structure. These two practices, which are generally viewed as

unethical at best and even illegal, create an ideal ground for analyzing the potential

relation between the likelihood to deceive and organizational structure. Our results show

that team-managed funds are less likely to involve in these dishonest fund performance

enhancing activities. Moreover, in some instances, such as window dressing, based on

selling poorly-performed stocks at the end of quarters and the year, we are unable to fund

any significant evidence of such behavior among team-managed funds. Across all our

tests, we document a negative relation between the extent of the evidence of the two

trading tactics and team size. These cross-managerial structure results hold irrespective

of additional controls related to such fund characteristics as fund returns, size, and

turnover that could also affect the actual propensity of fund manager to deceive. We also

show that portfolio pumping activity, which is again greatly reduced among team-

managed funds, including the best performers, is present most profoundly among single-

150

managed funds showing the worst prior performance. Thus, our findings provide novel

empirical support for the benefits of team-management in the fund industry.

151

Appendix

A.1: SEC cases related to portfolio pumping

Case 1: Excerpted from Litigation Release No. 20046 / March 16, 200750

SEC v. Burton G. Friedlander et al., Civil Action No. 01 Civ. 4683 (KMW) (S.D.N.Y.)

On February 21, 2007, United States District Judge Kimba Wood entered final judgments by

consent against Burton Friedlander and four entities he formerly controlled. These final judgments

conclude the U.S. Securities and Exchange Commission's action, except for a final distribution by the

court-appointed receiver.

The Commission filed its original complaint in May 2001, alleging fraud in connection with

Friedlander’s management of the assets of Friedlander International Limited, an overseas hedge fund. The

Commission alleged that Friedlander inflated the hedge fund’s net asset value by improperly and arbitrarily

valuing certain unlisted securities of a company in which Friedlander and entities he controlled had heavily

invested. The Commission’s complaint also alleged that Friedlander engaged in “portfolio pumping” by

purchasing a thinly-traded common stock as part of a manipulative scheme to inflate the value of that stock

and to inflate the hedge fund's net asset value…

Case 2: Excerpted from Litigation Release No. 21865 / February 25, 201151

SEC v. Todd M. Ficeto, Florian Homm, Colin Heatherington, Hunter World Markets, Inc., and Hunter

Advisors, LLC et al., Case No. CV-11-1637 GHK (RZx) (C.D. Cal. February 24, 2011)

The Securities and Exchange Commission charged two securities professionals, a hedge fund

trader, and two firms involved in a scheme that manipulated several U.S. microcap stocks and generated

more than $63 million in illicit proceeds through stock sales, commissions and sales credits.

According to the SEC’s complaint filed in the U.S. District Court for the Central District of

California, Homm along with Ficeto and Heatherington conducted the scheme from September 2005 to

50 See details at http://sec.gov/litigation/litreleases/2007/lr20046.htm. 51 See details at http://sec.gov/litigation/litreleases/2011/lr21865.htm.

152

September 2007… The SEC alleges that Florian Homm of Spain and Todd M. Ficeto of Malibu, Calif.,

conducted the scheme through their Beverly Hills, Calif.-based broker-dealer Hunter World Markets Inc.

(HWM) with the assistance of Homm’s close associate Colin Heatherington, a trader who lives in Canada.

They brought microcap companies public through reverse mergers and manipulated upwards the stock

prices of these thinly-traded stocks before selling their shares at inflated prices to eight offshore hedge

funds controlled by Homm. Their manipulation of the stock prices allowed Homm to materially overstate

by at least $440 million the hedge funds’ performance and net asset values (NAVs) in a fraudulent practice

known as “portfolio pumping…”

A.2: SEC cases related to window dressing

Case 1: Excerpted from Litigation Release No. 19170 / April 6, 200552

SEC v. Jeff Thomas Allen et al., Civil Action No. 05-453 (W.D. Pa.)

The Securities and Exchange Commission (“Commission”) announced that on April 6, 2005, it

filed a civil action in the United States District Court for the Western District of Pennsylvania against Jeff

Thomas Allen, of Pittsburgh, Pennsylvania, and James Barlow Smith, of Saxonburg, Pennsylvania. Allen

was the President, CEO, Chief Investment Officer and majority shareholder of Advanced Investment

Management, Inc. (“AIM”), a now-defunct investment adviser previously registered with the Commission.

Smith was AIM’s Vice President of Equity Trading...

The Commission’s Complaint alleges that AIM’s investment strategy involved matching or

exceeding the performance of the S&P 500 Index through the use of derivatives rather than direct

investment in the equities that comprised the Index… The Complaint further alleges that, from at least

January 2002 through July 2002, Allen and Smith conducted unauthorized trading in numerous client

accounts, and in violation of advisory agreements. In particular, from April through July 2002, during a

time when the S&P 500 Index dropped almost 29 percent, the defendants improperly increased market

exposure in an effort to recover from past losses. This trading caused market exposure in some accounts to

52 See details at http://sec.gov/litigation/litreleases/lr19170.htm.

153

reach levels as high as 500 percent, which, in turn, caused more than $415 million in client losses. In order

to conceal the effect of their trading, which otherwise would have been disclosed in monthly account

statements, Allen and Smith sold the unauthorized positions before month-end, and repurchased them

shortly thereafter. This strategy of “window dressing” prevented clients from discovering the scheme…

154

Table 1 Summary Statistics of Daily Fund Returns and Fund Characteristics

Team Size

Overall Single Team 2FM 3FM 4FM 5+FM Daily Excess Return Mean 0.010 0.009 0.011 0.011 0.012 0.011 0.011 SD 0.567 0.585 0.553 0.561 0.574 0.532 0.511 Investment Objectives:

Aggressive Growth Mean 0.017 0.016 0.018 0.018 0.018 0.020 0.017 SD 0.721 0.741 0.706 0.720 0.729 0.681 0.636

Growth Mean 0.010 0.009 0.010 0.010 0.011 0.008 0.011 SD 0.547 0.569 0.531 0.537 0.551 0.504 0.502

Growth & Income Mean 0.005 0.003 0.007 0.006 0.007 0.012 0.007 SD 0.419 0.422 0.417 0.417 0.424 0.444 0.389

Equity Income Mean 0.005 0.004 0.006 0.004 0.010 0.007 0.008 SD 0.473 0.483 0.462 0.484 0.451 0.404 0.438

Obs. 7,053 3,013 4,040 1,988 939 459 652 Fund Size Mean 1,080 1,040 1,110 745 931 917 2,620 SD 4,600 4,070 4,960 2,090 2,590 2,350 11,100 Fund Turnover Mean 0.900 0.961 0.856 0.859 0.903 0.830 0.796 SD 1.076 1.383 0.774 0.767 0.915 0.660 0.632

This table reports the mean and standard deviation (SD) of daily returns of domestic equity mutual funds in the U.S. across various managerial structures and investment objectives from January 2, 1992 to December 31, 2010. Based on the number of fund managers listed in Morningstar Direct database, funds are categorized into two broad managerial structures: Single and Team. Funds with only one fund manager are classified as Single whereas funds with more than one manager are classified as Team. Team Size represents funds with two, three, four and five (or more) fund managers. Daily Excess Return is the daily excess fund return (%) computed as the difference between the daily net fund return and the daily S&P 500 index return. Fund Size (millions, $) is the total net assets under the management of the fund at the end of the year. Fund Turnover is the minimum of aggregated sales or aggregated purchases of securities in a year divided by the average 12-month total net assets of the fund. The four fund investment objectives are Aggressive Growth, Growth, Growth & Income, and Equity Income. The number of observations is reported in 1000s.

155

Table 2 Effect of Managerial Structure on Portfolio Pumping Activity

Team Size

Single Team 2FM 3FM 4FM 5+FM YEND 0.323 -0.106 -0.081 -0.107 -0.139 -0.162 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) YBEG -0.259 0.043 0.021 0.025 0.073 0.114 (0.000) (0.000) (0.077) (0.094) (0.000) (0.000) QEND 0.202 -0.054 -0.037 -0.047 -0.082 -0.098 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) QBEG -0.213 0.036 0.022 0.021 0.052 0.087 (0.000) (0.000) (0.007) (0.067) (0.000) (0.000)

This table reports panel regression coefficients of daily excess fund returns on managerial structure using U.S. domestic equity mutual fund data from January 2, 1992 to December 31, 2010. The regression model is:

.,13121110987

65

43210,

tiitttttt

itit

ititititti

eTeambMBEGbMENDbQBEGbQENDbYBEGbYENDb

TeamMBEGbTeamMENDb

TeamQBEGbTeamQENDbTeamYBEGbTeamYENDbbr

++++++++

+×+×+

+×+×+×+×+=

The dependent variable is the daily fund return (net of expenses) in excess of daily S&P500 index return. Independent variables include Team, defined as a dummy variable which equals one if the fund has two (or more) fund managers and zero otherwise; YEND – the last trading day of year dummy; YBEG – the first trading day of the year dummy; QEND – the last trading day of the quarter, that is, March, June or September dummy; QBEG – the first trading day of the quarter, that is, April, July or October dummy; MEND – the last trading day of February, April, May, July, August, October or November dummy; and MBEG – the first trading day of February, March, May, June, August, September, November or December dummy. The coefficients reported in the table are: the interaction terms of Team and YEND, b1, YBEG, b2, QEND, b3, and QBEG, b4. The standard errors are clustered by fund. The p-values for the test if the estimates are different from zero are in parentheses.

156

Table 3 Effect of Managerial Structure on Portfolio Pumping across Investment Objectives

Team Size

Single Team 2FM 3FM 4FM 5+FM Aggressive Growth YEND 0.496 -0.155 -0.117 -0.171 -0.212 -0.223 (0.000) (0.000) (0.002) (0.000) (0.000) (0.000) YBEG -0.489 0.068 0.035 0.048 0.131 0.169 (0.000) (0.004) (0.223) (0.167) (0.002) (0.000) QEND 0.400 -0.130 -0.078 -0.127 -0.231 -0.234 (0.000) (0.000) (0.002) (0.000) (0.000) (0.000) QBEG -0.413 0.092 0.068 0.062 0.126 0.198 (0.000) (0.000) (0.001) (0.039) (0.000) (0.000) Growth YEND 0.289 -0.113 -0.088 -0.115 -0.144 -0.162 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) YBEG -0.231 0.044 0.020 0.042 0.069 0.104 (0.000) (0.001) (0.191) (0.028) (0.001) (0.000) QEND 0.168 -0.045 -0.032 -0.042 -0.062 -0.079 (0.000) (0.000) (0.001) (0.000) (0.000) (0.000) QBEG -0.206 0.031 0.017 0.025 0.051 0.069 (0.000) (0.000) (0.069) (0.058) (0.002) (0.000) Growth & Income YEND 0.235 -0.022 -0.012 -0.003 -0.015 -0.074 (0.000) (0.260) (0.569) (0.909) (0.727) (0.024) YBEG -0.116 0.030 0.041 -0.028 0.033 0.067 (0.000) (0.053) (0.031) (0.237) (0.342) (0.021) QEND 0.095 0.005 -0.005 0.025 0.023 -0.001 (0.000) (0.603) (0.676) (0.105) (0.338) (0.943) QBEG -0.043 0.003 0.003 -0.015 -0.009 0.031 (0.000) (0.744) (0.784) (0.317) (0.774) (0.061) Equity Income YEND 0.292 -0.082 -0.060 -0.087 -0.132 -0.130 (0.000) (0.005) (0.084) (0.031) (0.012) (0.027) YBEG -0.100 -0.042 -0.070 -0.037 -0.019 0.083 (0.000) (0.147) (0.058) (0.420) (0.737) (0.190) QEND 0.145 -0.038 -0.031 -0.038 -0.057 -0.052 (0.000) (0.019) (0.094) (0.099) (0.100) (0.136) QBEG -0.006 -0.014 -0.023 0.000 -0.021 0.015 (0.698) (0.510) (0.366) (0.993) (0.552) (0.749)

This table reports panel regression coefficients of daily excess fund returns on managerial structure across different investment objectives using the US domestic equity mutual fund data from January 2, 1992 to December 31, 2010. The regression specification is as in Table 2, but it is rerun separately for each of the four investment objectives: Aggressive Growth, Growth, Growth & Income, and Equity Income. All variables are defined and reported as in Table 2. The p-values for the test if the estimates are different from zero are in parentheses.

157

Table 4 Effect of Fund Size on Portfolio Pumping and Managerial Structure Relation

Team Size

Single Team 2FM 3FM 4FM 5+FM Quartile 1 (Smallest) YEND 0.369 -0.160 -0.145 -0.105 -0.233 -0.255 (0.000) (0.000) (0.000) (0.003) (0.000) (0.000) YBEG -0.218 0.004 0.007 -0.037 -0.023 0.081 (0.000) (0.866) (0.782) (0.263) (0.530) (0.016) QEND 0.195 -0.063 -0.052 -0.040 -0.111 -0.104 (0.000) (0.000) (0.001) (0.050) (0.000) (0.000) QBEG -0.216 0.030 0.025 -0.006 0.063 0.084 (0.000) (0.021) (0.108) (0.767) (0.015) (0.000) Quartile 2 YEND 0.332 -0.098 -0.053 -0.136 -0.140 -0.168 (0.000) (0.000) (0.038) (0.000) (0.000) (0.000) YBEG -0.280 0.055 0.022 0.056 0.124 0.114 (0.000) (0.003) (0.297) (0.037) (0.000) (0.000) QEND 0.221 -0.080 -0.052 -0.062 -0.149 -0.154 (0.000) (0.000) (0.001) (0.001) (0.000) (0.000) QBEG -0.206 0.032 0.015 0.024 0.041 0.104 (0.000) (0.015) (0.351) (0.235) (0.103) (0.000) Quartile 3 YEND 0.325 -0.108 -0.095 -0.124 -0.102 -0.131 (0.000) (0.000) (0.000) (0.000) (0.006) (0.000) YBEG -0.316 0.090 0.071 0.066 0.136 0.154 (0.000) (0.000) (0.002) (0.018) (0.000) (0.000) QEND 0.226 -0.065 -0.043 -0.073 -0.076 -0.115 (0.000) (0.000) (0.007) (0.000) (0.003) (0.000) QBEG -0.234 0.051 0.032 0.043 0.057 0.120 (0.000) (0.000) (0.047) (0.050) (0.024) (0.000) Quartile 4 (Largest) YEND 0.267 -0.060 -0.032 -0.064 -0.087 -0.103 (0.000) (0.002) (0.145) (0.021) (0.015) (0.002) YBEG -0.221 0.018 -0.015 0.006 0.040 0.096 (0.000) (0.327) (0.498) (0.828) (0.199) (0.002) QEND 0.167 -0.012 -0.003 -0.014 -0.002 -0.037 (0.000) (0.295) (0.828) (0.441) (0.930) (0.037) QBEG -0.197 0.028 0.019 0.020 0.048 0.049 (0.000) (0.035) (0.231) (0.341) (0.062) (0.025)

This table reports panel regression coefficients of daily excess fund returns on managerial structure across different fund size quartiles using U.S. domestic equity mutual fund data from January 2, 1992 to December 31, 2010. The regression specification is as in Table 2, but it is rerun separately for each fund size quartile. All variables are defined and reported as in Table 2. The p-values for the test if the estimates are different from zero are in parentheses.

158

Table 5 Effect of Fund Turnover on Portfolio Pumping and Managerial Structure Relation

Team Size

Single Team 2FM 3FM 4FM 5+FM Quartile 1 (Lowest) YEND 0.270 -0.070 -0.064 -0.058 -0.128 -0.063 (0.000) (0.000) (0.005) (0.041) (0.000) (0.097) YBEG -0.181 0.018 -0.016 -0.006 0.046 0.138 (0.000) (0.291) (0.444) (0.815) (0.139) (0.000) QEND 0.149 -0.038 -0.019 -0.032 -0.074 -0.076 (0.000) (0.001) (0.158) (0.076) (0.002) (0.000) QBEG -0.145 0.016 0.009 0.021 0.005 0.039 (0.000) (0.176) (0.507) (0.316) (0.842) (0.037) Quartile 2 YEND 0.298 -0.076 -0.058 -0.065 -0.123 -0.116 (0.000) (0.000) (0.010) (0.010) (0.000) (0.000) YBEG -0.220 0.027 0.016 0.007 0.066 0.067 (0.000) (0.137) (0.481) (0.784) (0.032) (0.029) QEND 0.182 -0.047 -0.033 -0.046 -0.067 -0.075 (0.000) (0.000) (0.019) (0.007) (0.002) (0.000) QBEG -0.189 0.046 0.033 0.026 0.072 0.094 (0.000) (0.000) (0.020) (0.120) (0.001) (0.000) Quartile 3 YEND 0.327 -0.132 -0.092 -0.154 -0.149 -0.206 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) YBEG -0.283 0.042 0.025 0.014 0.048 0.119 (0.000) (0.026) (0.270) (0.642) (0.181) (0.000) QEND 0.209 -0.051 -0.039 -0.054 -0.039 -0.085 (0.000) (0.000) (0.007) (0.004) (0.140) (0.000) QBEG -0.224 0.037 0.019 0.022 0.079 0.079 (0.000) (0.005) (0.214) (0.268) (0.004) (0.000) Quartile 4 (Highest) YEND 0.328 -0.131 -0.103 -0.152 -0.139 -0.184 (0.000) (0.000) (0.000) (0.000) (0.002) (0.000) YBEG -0.338 0.077 0.046 0.088 0.134 0.132 (0.000) (0.000) (0.061) (0.009) (0.001) (0.001) QEND 0.244 -0.079 -0.054 -0.053 -0.160 -0.148 (0.000) (0.000) (0.001) (0.010) (0.000) (0.000) QBEG -0.281 0.046 0.039 0.000 0.060 0.137 (0.000) (0.002) (0.023) (0.990) (0.023) (0.000)

This table reports panel regression coefficients of daily excess fund returns on managerial structure across different fund turnover quartiles using U.S. domestic equity mutual fund data from January 2, 1992 to December 31, 2010. The regression specification is as in Table 2, but it is rerun separately for each fund turnover quartile. All variables are defined and reported as in Table 2. The p-values for the test if the estimates are different from zero are in parentheses.

159

Table 6 Summary Statistics of Trading Intensity across Funds with Different Managerial Structures

Team Size Overall Single Team 2FM 3FM 4FM 5+FM

Selling Intensity: Extreme Losers Mean 1.286 1.304 1.273 1.288 1.289 1.280 1.207

SD 0.939 0.953 0.928 0.943 0.939 0.934 0.867 Extreme Winners Mean 1.449 1.440 1.455 1.466 1.451 1.455 1.429

SD 0.889 0.892 0.887 0.886 0.899 0.902 0.862 Obs. 240 97 142 69 31 16 25

Buying Intensity: Extreme Losers Mean 1.423 1.428 1.419 1.453 1.386 1.373 1.399

SD 1.161 1.168 1.156 1.187 1.122 1.142 1.117 Extreme Winners Mean 1.420 1.437 1.408 1.427 1.419 1.420 1.325

SD 1.077 1.107 1.056 1.086 1.056 1.032 0.973 Obs. 293 121 171 84 38 19 28

Fund Excess Return Mean 0.744 0.694 0.779 0.798 0.780 0.726 0.760 SD 5.193 5.417 5.030 5.122 5.448 4.692 4.338

Fund Size Mean 702 737 677 468 642 528 1469 SD 2,747 2,615 2,836 1,413 1,781 1,376 6,080

Fund Turnover Mean 0.884 0.948 0.836 0.836 0.894 0.812 0.773 SD 1.077 1.398 0.751 0.732 0.911 0.630 0.619

This table reports the mean and standard deviation (SD) of trading intensity among U.S. domestic equity mutual funds across various managerial structures from January 1992 to December 2010. Based on the number of fund managers listed in Morningstar Direct database, funds are categorized into two broad managerial structures: Single and Team. Funds with only one fund manager are classified as Single whereas funds with more than one manager are classified as Team. Team Size represents funds with two, three, four and five (or more) fund managers. Selling Intensity is defined as

( ) ( )( ) ( ) ( )( ),,1,,,,1,,, ∑∑ −−ii

kqiHOLDkqiSELLkqiHOLDkqiSELL

where SELL(i,q,k) is the dollar value of sales by fund k in quarter q and performance group i, and HOLD (i,q-1,k) is the dollar value of holdings at the end of the quarter q-1 of the exact same stocks as those in performance group i in quarter q. Buying Intensity is defined as

( ) ( )( ) ( ) ( )( ),,.,.,,,, ∑∑ iiqiHOLDUNIVqiHOLDUNIVkqiBUYkqiBUY

where BUY(i,q,k) is the dollar purchases by fund k in performance group i in quarter q and UNIV.HOLD (i,q) is the value of CRSP universe holding in quarter q in performance group i. Extreme Losers is defined as the lowest stock performance quintile each quarter, where stocks have the lowest returns over the past year up to the end of that quarter. Extreme Winners is defined as the highest performance quintile each quarter, where stocks have the highest stock returns over the past year up to the end of that quarter. These definitions follow Lakonishok et al. (1991). Fund Excess Return is the quarterly excess fund return (%) computed as the difference between the quarterly net fund return and the daily S&P 500 index return. Fund Size (millions, $) is the total net assets under the management of the fund at the end of the year. Fund Turnover is the minimum of aggregated sales or aggregated purchases of securities of the year divided by the average 12-month total net assets of the fund. The number of observations is reported in 1000s.

160

Table 7 Effect of Managerial Structure on Window Dressing Activity

Selling Intensity Buying Intensity Q1-Q3 Q4 Diff(Q4-13) Q1-Q3 Q4 Diff(Q4-13)

Extreme Losers: Overall 1.282 1.290 0.007 1.417 1.422 0.005 0.005 0.009 (0.465) 0.005 0.009 (0.625) Single 1.296 1.327 0.031 1.426 1.434 0.008 0.008 0.014 (0.059) 0.009 0.015 (0.646) Team 1.276 1.263 -0.013 1.418 1.424 0.006 0.007 0.011 (0.324) 0.007 0.013 (0.693) Diff(Team-Single) -0.019 -0.064 -0.008 -0.010

(0.075) (0.001) (0.483) (0.596) Extreme Winners:

Overall 1.450 1.440 -0.009 1.437 1.398 -0.039 0.004 0.007 (0.276) 0.005 0.009 (0.001) Single 1.445 1.427 -0.018 1.451 1.399 -0.052 0.007 0.012 (0.193) 0.009 0.014 (0.002) Team 1.456 1.452 -0.004 1.416 1.383 -0.034 0.006 0.010 (0.735) 0.007 0.011 (0.012) Diff(Team-Single) 0.011 0.025 -0.034 -0.016 (0.254) (0.103) (0.002) (0.378) This table shows the effect of managerial structure on window dressing of U.S. domestic equity mutual funds from January 1992 to December 2010. It reports the average selling and buying intensity of stocks over funds and year comparing across the first three quarters and the last quarter of the year. Funds are categorized into two broad managerial structures: Single, which are funds with only one fund manager and Team, which are funds with two or more managers. Selling Intensity is defined as the ratio of stocks sold in a performance group by a fund to holdings of the same stocks at the end of the previous quarter divided by the ratio of total stocks sold in this quarter to holdings at the end of the previous quarter. Buying Intensity is defined as the fraction of stock purchases in a performance group by a fund relative to the fraction of the universe holdings in that performance group. Extreme Winners and Extreme Losers are defined as in Table 6. The standard errors are shown below the mean estimates. Diff(Q4-13) is the difference between quarter 4 and quarters 1-3 across selling and buying intensities, respectively. Diff(Team-Single) is the difference of means between single and team-managed funds within quarters 1-3 and quarter 4. The standard errors are below intensity estimates. The p-values for the difference tests are in parentheses.

161

Table 8 Window Dressing across Teams of Different Sizes

Panel A: Window dressing across team sizes

Selling Intensity (Extreme Losers) Buying Intensity (Extreme Winners) Q1-Q3 Q4 Diff(Q4-13) Q1-Q3 Q4 Diff(Q4-13)

Single 1.296 1.327 0.031 1.426 1.434 0.008 0.008 0.014 (0.059) 0.009 0.015 (0.646)

2FM 1.295 1.269 -0.026 1.446 1.469 0.023 0.010 0.017 (0.178) 0.011 0.019 (0.277)

3FM 1.289 1.290 0.001 1.392 1.371 -0.021 0.015 0.025 (0.982) 0.015 0.025 (0.477)

4FM 1.274 1.298 0.024 1.375 1.366 -0.010 0.020 0.034 (0.547) 0.021 0.037 (0.815)

5+FM 1.212 1.193 -0.019 1.399 1.399 0.001 0.015 0.025 (0.522) 0.017 0.029 (0.987)

Panel B: Difference in selling and buying intensities across management structures

Selling Intensity (Extreme Losers) Buying Intensity (Extreme Winners) Difference Q1-Q3 Q4 Q1-Q3 Q4

2FM - Single -0.001 -0.058 0.020 0.036 (0.966) (0.009) (0.142) (0.131)

3FM - Single -0.007 -0.037 -0.034 -0.063 (0.696) (0.205) (0.055) (0.034)

4FM - Single -0.021 -0.028 -0.051 -0.068 (0.334) (0.461) (0.033) (0.090)

5+FM - Single -0.084 -0.134 -0.027 -0.035 (0.000) (0.000) (0.172) (0.304)

This table shows window dressing activity of mutual funds managed by a single manager as well as by teams of two, three, four, and five (or more) managers using U.S. domestic equity mutual funds from January 1992 to December 2010. Panel A reports equal-weighted average selling and buying intensity of extreme winner stocks and extreme loser stocks across the first three quarters and the fourth quarter of the year. The standard errors are shown below the mean estimates. Panel B reports the difference in means between funds with two, three, four, and five (or more) fund managers and single-manager funds within quarters 1-3 and quarter 4. Single is defined as funds with only one fund manager; 2FM, 3FM, 4FM and 5+FM are funds with two, three, four, and five (or more) fund managers. Other measures are defined as in Tables 6 and 7. Diff(Q4-13) is the difference between quarter 4 and quarters 1-3 across selling and buying intensities, respectively. The standard errors are below intensity estimates. The p-values for the difference tests are in parentheses.

162

Table 9 Managerial Structure and Window Dressing across Investment Objectives

Panel A: Selling intensity in extreme loser stocks Aggressive Growth Growth Growth & Income Equity Income Q1-Q3 Q4 Diff(Q4-13) Q1-Q3 Q4 Diff(Q4-13) Q1-Q3 Q4 Diff(Q4-13) Q1-Q3 Q4 Diff(Q4-13)

Overall 1.176 1.168 -0.009 1.314 1.328 0.014 1.299 1.310 0.011 1.443 1.438 -0.005 0.010 0.017 (0.658) 0.007 0.012 (0.287) 0.014 0.023 (0.675) 0.031 0.051 (0.933)

Single 1.195 1.200 0.005 1.306 1.343 0.037 1.339 1.379 0.040 1.528 1.583 0.055 0.017 0.029 (0.880) 0.011 0.018 (0.082) 0.024 0.041 (0.388) 0.047 0.075 (0.536)

Team 1.155 1.138 -0.022 1.315 1.309 -0.005 1.295 1.274 -0.021 1.384 1.262 -0.122 0.013 0.022 (0.507) 0.009 0.015 (0.772) 0.019 0.030 (0.547) 0.043 0.071 (0.151)

Diff(Team-Single) -0.040 -0.062 0.008 -0.034 -0.044 -0.105 -0.144 -0.321 (0.063) (0.082) (0.552) (0.159) (0.144) (0.035) (0.025) (0.002)

Panel B: Buying intensity in extreme winner stocks

Aggressive Growth Growth Growth & Income Equity Income Q1-Q3 Q4 Diff(Q4-13) Q1-Q3 Q4 Diff(Q4-13) Q1-Q3 Q4 Diff(Q4-13) Q1-Q3 Q4 Diff(Q4-13)

Overall 1.811 1.762 -0.049 1.431 1.402 -0.030 0.964 0.941 -0.023 0.879 0.847 -0.032 0.012 0.020 (0.042) 0.006 0.011 (0.018) 0.010 0.017 (0.240) 0.021 0.033 (0.422)

Single 1.855 1.809 -0.045 1.437 1.371 -0.066 0.963 0.944 -0.020 0.874 0.860 -0.014 0.021 0.034 (0.263) 0.011 0.017 (0.001) 0.017 0.029 (0.555) 0.029 0.050 (0.804)

Team 1.766 1.698 -0.068 1.421 1.402 -0.018 0.928 0.901 -0.028 0.833 0.772 -0.062 0.016 0.026 (0.029) 0.009 0.014 (0.274) 0.013 0.021 (0.275) 0.029 0.042 (0.263)

Diff(Team-Single) -0.089 -0.111 -0.017 0.031 -0.035 -0.043 -0.040 -0.088 (0.001) (0.008) (0.221) (0.156) (0.100) (0.216) (0.322) (0.181)

This table reports the average selling and buying intensities of funds within different investment objectives comparing across quarters 1-3 and quarter 4. The sample includes U.S. domestic equity mutual funds from January 1992 to December 2010. Investment objectives are: Aggressive Growth, Growth, Growth & Income, and Equity Income. Panel A reports the average selling intensity within extreme loser stocks across quarters 1-3 and quarter 4. Panel B reports the average buying intensity within extreme winner stocks across quarters 1-3 and quarter 4. Other measures are defined as in Tables 6 and 7. The standard errors are shown below the mean estimates. Diff(Q4-13) is the difference between quarter 4 and quarters 1-3 across selling and buying intensities, respectively. Diff(Team-Single) is the difference of means between single and team-managed funds within quarters 1-3 and quarter 4. The standard errors are below intensity estimates. The p-values for the difference tests are in parentheses.

163

Table 10 Effect of Fund Size and Turnover on Window Dressing and Managerial Structure Relation

Panel A: Results by fund size

Selling Intensity (Extreme Losers) Buying Intensity (Extreme Winners) Q1-Q3 Q4 Diff(Q4-13) Q1-Q3 Q4 Diff(Q4-13)

Large Funds: Single 1.227 1.277 0.051 1.315 1.311 -0.004 0.014 0.025 (0.072) 0.015 0.025 (0.884) Team 1.235 1.262 0.027 1.254 1.226 -0.028 0.013 0.022 (0.274) 0.012 0.020 (0.235)

Diff(Team-Single) 0.008 -0.016 -0.060 -0.084 (0.678) (0.638) (0.002) (0.007)

Small Funds: Single 1.353 1.384 0.031 1.515 1.482 -0.033 0.020 0.034 (0.432) 0.019 0.031 (0.366) Team 1.342 1.287 -0.055 1.445 1.427 -0.018 0.016 0.026 (0.077) 0.014 0.025 (0.535)

Diff(Team-Single) 0.011 0.097 -0.071 -0.055 (0.663) (0.020) (0.003) (0.163)

Panel B: Results by fund turnover

Selling Intensity (Extreme Losers) Buying Intensity (Extreme Winners) Q1-Q3 Q4 Diff(Q4-13) Q1-Q3 Q4 Diff(Q4-13)

High Turnover Funds: Single 1.187 1.203 0.016 1.795 1.732 -0.063 0.014 0.024 (0.545) 0.017 0.028 (0.056) Team 1.223 1.232 0.009 1.799 1.762 -0.037 0.013 0.022 (0.709) 0.014 0.024 (0.172)

Diff(Team-Single) 0.037 0.030 0.004 0.030 (0.051) (0.361) (0.854) (0.415)

Low Turnover Funds: Single 1.430 1.463 0.032 1.190 1.129 -0.061 0.022 0.036 (0.445) 0.018 0.029 (0.082) Team 1.413 1.376 -0.037 1.079 1.043 -0.037 0.019 0.031 (0.312) 0.015 0.023 (0.195)

Diff(Team-Single) -0.017 -0.086 -0.111 -0.087 (0.547) (0.066) (0.001) (0.018)

This table compares average selling and buying intensities of single- and team-managed funds across quarters 1-3 and quarter 4 controlling for fund size (Panel A) and fund turnover (Panel B). The sample includes U.S. domestic equity mutual funds from January 1992 to December 2010. At the end of each quarter, funds are sorted into size quartiles based on the total net assets under their management. The top (bottom) quartiles are classified as Large (Small) Funds. Also, at the end of each quarter, funds are sorted into turnover quartiles based on the annual fund turnover. The top (bottom) quartiles are classified as High (Low) Turnover Funds. Other measures are defined as in Tables 6 and 7. The standard errors are shown below the mean estimates. Diff(Q4-13) is the difference between quarter 4 and quarters 1-3 across selling and buying intensities, respectively. Diff(Team-Single) is the difference of means between single and team-managed funds within quarters 1-3 and quarter 4. The standard errors are below intensity estimates. The p-values for the difference tests are in parentheses.

164

Table 11 Effect of Fund Performance on Window Dressing and Managerial Structure Relation

Selling Intensity (Extreme Losers) Buying Intensity (Extreme Winners) Q1-Q3 Q4 Diff(Q4-13) Q1-Q3 Q4 Diff(Q4-13)

Winner Funds: Single 1.302 1.323 0.021 1.677 1.626 -0.051 0.019 0.031) (0.567) 0.018) 0.029 (0.134) Team 1.298 1.342 0.043 1.627 1.600 -0.026 0.015 0.028 (0.161) 0.014) 0.024 (0.346)

Diff(Team-Single) -0.004 0.019 -0.051 -0.026 (0.880) (0.653) (0.026) (0.491)

Quartile 3: Single 1.285 1.360 0.076 1.370 1.309 -0.062 0.018 0.030 (0.025) 0.017 0.027 (0.052) Team 1.273 1.263 -0.010 1.352 1.311 -0.041 0.014 0.023 (0.711) 0.013 0.021 (0.102)

Diff(Team-Single) -0.012 -0.098 -0.018 0.002 (0.589) (0.009) (0.384) (0.948)

Quartile 2: Single 1.324 1.377 0.054 1.278 1.266 -0.012 0.017 0.028 (0.093) 0.016 0.025 (0.694) Team 1.292 1.246 -0.047 1.301 1.221 -0.080 0.014 0.022 (0.074) 0.020 0.020 (0.001)

Diff(Team-Single) -0.031 -0.132 0.023 -0.045 (0.144) (0.001) (0.25) (0.155)

Loser Funds: Single 1.264 1.245 -0.019 1.428 1.365 -0.062 0.015 0.026 (0.529) 0.018 0.029 (0.080) Team 1.245 1.207 -0.037 1.361 1.364 0.002 0.013 0.020 (0.129) 0.014 0.025 (0.932)

Diff(Team-Single) -0.019 -0.038 -0.066 -0.002 (0.337) (0.244) (0.004) (0.963)

This table compares average selling and buying intensities of single- and team-managed funds across quarters 1-3 and quarter 4 controlling for different fund performance quartiles. The sample includes U.S. domestic equity mutual funds from January 1992 to December 2010. The Winner (Loser) funds are those within the top 25% (bottom 25%) of median performance across all funds in a given year. Fund performance is based on the net excess fund returns. Other measures are defined as in Tables 6 and 7. The standard errors are shown below the mean estimates. Diff(Q4-13) is the difference between quarter 4 and quarters 1-3 across selling and buying intensities, respectively. Diff(Team-Single) is the difference of means between single and team-managed funds within quarters 1-3 and quarter 4. The standard errors are below intensity estimates. The p-values for the difference tests are in parentheses.

165

Table 12 The Dot-Com Bubble: A Special Case of Window Dressing and Managerial Structure

Panel A: Window dressing during the Dot-Com bubble

Selling Intensity (Extreme Losers) Buying Intensity (Extreme Winners) Q1-Q3 Q4 Diff(Q4-13) Q1-Q3 Q4 Diff(Q4-13)

Overall 1.255 1.278 0.022 1.154 1.169 0.014 0.012 0.019 (0.310) 0.009 0.013 (0.38)

Single 1.261 1.323 0.062 1.147 1.174 0.027 0.017 0.028 (0.055) 0.013 0.020 (0.255)

Team 1.259 1.227 -0.032 1.144 1.137 -0.007 0.017 0.027 (0.334) 0.013 0.020 (0.756)

Diff(Team-Single) -0.003 -0.096 0.004 -0.003 (0.917) (0.013) (0.883) (0.182)

Panel B: Difference in selling and buying intensities across management structures during Dot-Com bubble

Selling Intensity (Extreme Losers) Buying Intensity (Extreme Winners) Difference Q1-Q3 Q4 Q1-Q3 Q4

2FM - Single 0.021 -0.049 0.006 -0.025 (0.479) (0.298) (0.777) (0.442)

3FM - Single -0.013 -0.102 0.067 0.041 (0.747) (0.116) (0.023) (0.363)

4FM - Single 0.024 -0.202 -0.049 -0.044 (0.698) (0.052) (0.287) (0.531)

5+FM - Single -0.118 -0.231 -0.188 -0.290 (0.032) (0.008) (0.000) (0.000)

This table shows the impact of managerial structure on window dressing activity of equity mutual funds during the Dot-com bubble. The sample includes U.S. domestic equity mutual funds from January 1996 to December 2000. Panel A reports equal-weighted average selling and buying intensities of single- and team-managed funds across quarters 1-3 and quarter 4. Panel B reports the difference in means between funds with two, three, four, and five (or more) fund managers and single-manager funds within quarters 1-3 and quarter 4. Other measures are defined as in Tables 6 and 7. Diff(Q4-13) is the difference between quarter 4 and quarters 1-3 across selling and buying intensities, respectively. Diff(Team-Single) is the difference of means between single and team-managed funds within quarters 1-3 and quarter 4. The standard errors are below intensity estimates. The p-values for the difference tests are in parentheses.

166

Figure 1. Fund Returns around the Year-end for Different Managerial Structures and Performance. The figure shows the relation between daily excess fund returns (in percent) on the last trading day of the year (upper half of the plot) as well as the first trading day of the year (lower half of the plot) and fund performance across all funds (dashed curve), single-managed funds (thick curve), and team-managed funds (thin curve). Fund performance is measures from the first trading day of the year to the second-to-last day of the same year and is split into 20 performance bins by 5% each. The sample covers the period between January 1, 1992 and December 31, 2010.

-0.5

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

0.4

0.5

5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100

Beg

inn

ing

of

year

Fu

nd

daily e

xcess r

etu

rn (

%)

E

nd

of

year

<< Worst Fund performance percentile Best >>

All funds Single-managed Team-managed

167

Figure 2. Fund Returns around the Year-end for Different Manager Team Sizes and Performance. The figure shows the relation between daily excess fund returns (in percent) on the last trading day of the year (upper half of the plot) as well as the first trading day of the year (lower half of the plot) and fund performance across single-managed funds (thick curve), funds with two or three managers (dashed curve), and funds with four, five and more managers (thin curve). The daily excess return data for funds with two or three managers as well as four, five or more managers are averaged. Fund performance is measures from the first trading day of the year to the second-to-last day of the same year and is split into 20 performance bins by 5% each. The sample covers the period between January 1, 1992 and December 31, 2010.

-0.5

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

0.4

0.5

5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100

Beg

inn

ing

of

year

Fu

nd

daily e

xcess r

etu

rn (

%)

E

nd

of

year

<< Worst Fund performance percentile Best >>

Single-managed 2-3 managers 4-5 managers

168

Chapter 5

Conclusions

169

This thesis comprises of three essays that use the U.S. equity mutual fund

industry to test three important economic questions. In doing so, this thesis makes four

key contributions to the literature. First, we shed new light on the importance of

investors' economic expectations in investment decisions and provide a plausible

explanation to how investor expectations enter investors' investment decisions. We show

that economic expectations positively influence investors' investment decisions and

investors' put more weight on their expectations particularly when fund-specific

information is noisy and less reliable. Second, we document a large discrepancy in

managerial structure reporting in CRSP database. We find that CRSP database

inaccurately reports the number of portfolio managers responsible for day to day

activities of a fund. The discrepancy affects, on average, one-fifth of the sample per year.

Once we correct this discrepancy, we resolve the conflicting evidence in the existing

literature on the benefits of team compared to single managers. Third, we provide new

empirical evidence on benefits of teamwork conditional of team size, diversity and

geographical location. Our results not only show that team-managed funds outperform

single-managed funds but also find that the outperformance is the highest among three-

member teams, teams with homogeneous members and teams located in financial

centers. Lastly, we are the first to empirically show that team-based managerial structure

can significantly reduce incentives of portfolio manager to cheat or engage in deceptive

behavior. Our results strongly suggest that team-managed funds are less likely to involve

in two deceptive trading practices, namely, portfolio pumping and window dressing. We

also document that exists a negative relation between team size and likelihood of

deception/cheating.

We now elaborate of the findings of each of the three essays in detail. In the first

essay, we explore the importance of investors' expectations about future economic

growth in their investment decisions. We also provide a plausible explanation to how

170

investor expectations might enter their investment decisions. We start by constructing a

new measure of investors' economic expectations (INVEXP) that captures what investors

think will happen to the overall economy in a year. INVEXP offers three distinct

advantages over existing proxies. First, it is a direct and cleaner measure because it uses

survey data from individual households and concentrates only on questions related to

investors' expectation about economy. Second, it is highly correlated with known state

variables which predict future economic growth. And third, it has desirable statistical

properties such as relative stability, stationarity and lines up very well with the known

historical periods of optimism and pessimism. Then, we investigate whether INVEXP

affects mutual fund flows. We find that INVEXP positively relates to mutual fund flows.

This result suggests that investors increase their allocation to riskier equity fund when

they are optimistic about the future. This effect is robust to inclusion of various fund

characteristics, such as performance and fees, which are known in the literature to affect

mutual fund flows as well as various macroeconomic variables, such as stock market

returns and interest rates. Then, we investigate the mechanism through which

expectations enter investors' investment decision-making process. Financial theory

argues that investors trade-off different pieces of information based on the quality of

information while making investment decisions. Consistent with financial theory, we

find that investors put more weight on their expectations in making investment decisions

particular for those funds which have noisy and poor quality of fund-specific

information. Overall, our main conclusion is that economic expectations positively

influences investors' investment decisions and becomes particularly important to

investors whenever the fund-specific information is noisy and less informative.

The second essay examines whether teams are better than single managers at

managing mutual fund portfolios. In addressing this question, we first highlight a large

discrepancy in reported managerial structure between CRSP and MorningStar. We find

171

that CRSP database inaccurately reports the number of portfolio managers responsible

for day to day activities of a fund. This discrepancy ranges between as low as 10 percent

to as high as 26 percent of entire sample per year. Once we correct this discrepancy, we

find a positive (and statistically significant) relation between teams and fund

performance, unlike a negative relation documented by previous literature using CRSP

data. In the second part of this essay, we focus on conditional effects of teamwork based

on team size, location and team diversity. First, we examine the relation between team

size and fund performance. We observe a non-linear relation between team size and fund

performance. In particular, we find that three-member teams tend to generate highest

fund performance relative to single-manager funds. This result shows that not all teams

are equal and that the benefit of teamwork depends on the number of team members.

Second, we investigate whether location of a fund affects the team-fund performance

relation. We show that funds gain from team management only when they are located in

financial centers. We find no difference in teams’ performance relative to single

managers in non financial centers. This result confirms the intuition that teams gain

advantage over single managers because of their ability to access and collect a large

amount of private information abundant in financial centers. Then, we examine whether

diversity among team members affects fund performance. We show that funds with

more homogeneous team members in terms of age and educational background

outperform those with more heterogeneous managers. Further, we also analyze the effect

of teams on other fund characteristics such as fund flows, fees, size and turnover. We

find that funds managed by teams of portfolio managers receive significantly higher

flows relative to funds managed by single managers, ceteris paribus. This result suggests

that investors at large tend to prefer and trust teams relative to single managers. Overall,

our results strongly indicate that team-based managerial structure is beneficial, but its

magnitude of its benefits depends on team size, diversity and geographical location.

172

Finally, the third essay investigates whether team-based managerial structure reduces

likelihood of deception or cheating. We find that teams involve in significantly less

cheating than individual fund managers. In particular, we show that team-managed funds

involve significantly less in portfolio pumping, an illegal trading activity, and do not

involve at all in window dressing, a quasi-illegal and dishonest trading activity. Further,

we also document a negative relation between the extent of these two illegal and quasi-

illegal activities and team size. These results are robust to various fund characteristics

that are known to correlate with teams and team size such as fund size, fund portfolio

turnover and fund return. There can be several mechanisms at play behind this

phenomenon. But in our opinion, there are three mechanisms that seem most important.

First, teams increase cost of cheating by peer monitoring. Peer monitoring by other team

members increases the likelihood of getting caught which in turn reduce the likelihood of

cheating. Second, teams reduce benefits accruing from cheating to individuals since the

output of their joint production is divided among other non-cheating members. So the

entire cost of cheating is borne by an individual while the benefit of cheating is shared

among all team members. And third, individuals working within a team may experience

higher moral pressures such as guilt aversion. This might deter individuals within a team

to work against the interests of the team as a whole. Unfortunately, due to data

limitations, we are unable to differentiate across these mechanisms. Overall, our results

strongly suggest that team-based managerial structure can effectively reduce the

prevalence of illegal or quasi-illegal activities within an organization.

173

Bibliography

[1] Acemoglu, D., M. Kremer, and A. Mian, 2008, Incentives in markets, firms, and governments, Journal of Law, Economics, and Organization 24, 273-306.

[2] Adams, R., and D. Ferreira, 2010, Moderation in groups: Evidence from betting on ice break-ups in Alaska, Review of Economic Studies 77, 882-913.

[3] Agarwal, V., and L. Ma, 2011, Managerial multitasking in mutual fund industry, Working paper, Georgia State University.

[4] Agarwal, V., G. Gay, and L. Ling, 2011, Performance inconsistency in mutual funds: An investigation of window-dressing behavior, Working paper, Georgia State University.

[5] Alchian, A., and H. Demsetz, 1972, Production, information costs and economic organization, American Economic Review 62, 777-705.

[6] Allison, P., 1978, Measures of inequality, American Sociological Review 43, 865- 880.

[7] Angell, J., 1960, Uncertainty, Likelihoods and Investment Decisions, Quarterly

Journal of Economics 74 , 1-28.

[8] Arnott, R., and J. Stiglitz, 1991, Moral hazard and nonmarket institutions: Dysfunctional crowding out of peer monitoring?, American Economic Review 81, 179-190.

[9] Arrow, K., 1974, Limits to Organization, New York: Norton & Co.

[10] Bar, M., A. Kempf, and S. Ruenzi, 2011, Is a team different from the sum of its parts? Evidence from mutual fund managers, Review of Finance 15, 359-396.

[11] Barber, B., C. Heath, and T. Odean, 2003, Good reasons sell: Reason-based choice among group and individual investors in the stock market, Management

Science 49, 1636-1652.

[12] Barber, B., T. Odean, and L. Zheng, 2005, Out of Sight, Out of Mind: The Effects of Expenses on Mutual Fund Flows, Journal of Business 78, 2095-2119.

[13] Barry, C., and L. Starks, 1984, Investment management and risk sharing with multiple managers, Journal of Finance 39, 477-491.

[14] Becker, G., 1968, Crime and punishment: An economic approach, Journal of

Political Economy 76, 169-217.

174

[15] Ben-Rephael, A., S. Kandel, and A. Wohl, 2012, Measuring Investor Sentiment with Mutual Fund Flows, Journal of Financial Economics 104, 363-382.

[16] Berk, J., and R. Green, 2004, Mutual Fund Flows and Performance in Rational Markets, Journal of Political Economy 112, 1269-1295.

[17] Bernhardt, D., and R. Davies, 2005, Painting the tape: Aggregate evidence, Economic Letters 89, 306-311.

[18] Bhattacharyya, S., and V. Nanda, 2009, Portfolio pumping, trading activity and fund performance, Working paper, University of Michigan.

[19] Blinder, A., and J. Morgan, 2005, Are two heads better than one? Monetary policy by committee, Journal of Money, Credit and Banking 37, 789-811.

[20] Bliss, R., M. Porter, and C. Schwarz, 2008, Performance characteristics of individually-managed versus team-managed mutual funds, Journal of Portfolio

Management 34, 110-119.

[21] Bone, J., J. Hey, and J. Suckling, 1999, Are groups more (or less) consistent than individuals? Journal of Risk and Uncertainty 18, 63-81.

[22] Bornstein, G., and I. Yaniv, 1998, Individual and group behavior in the ultimatum game: Are groups more ‘rational’ players? Experimental Economics 1, 101-108.

[23] Brown, S., and W. Goetzmann, 1995, Performance Persistence, Journal of

Finance 50, 679-698.

[24] Cao, C., E. Chang, and Y. Wang, 2008, An Empirical Analysis of the Dynamic Relationship Between Mutual Fund Flow and Market Return Volatility, Journal

of Banking and Finance 32, 2111-2123.

[25] Capon, N., G. Fitzsimons, and R. Prince, 1996, An Individual Level Analysis of the Mutual Fund Investment Decision, Journal of Financial Services Research, 10, 59-82.

[26] Carhart, M., 1997, On persistence in mutual fund performance, Journal of

Finance 52, 57-82.

[27] Carhart, M., R. Kaniel, D. Musto, and A. Reed, 2002, Leaning for the tape: Evidence of gaming behavior in equity mutual funds, Journal of Finance 57, 661- 693.

[28] Carroll, C., 2003, Macroeconomic Expectations Of Households And Professional Forecasters, Quarterly Journal of Economics 118, 269-298.

175

[29] Chalmers, J., A. Kaul, and B. Phillips, 2012, The Wisdom of Crowds: Mutual

Fund Investors' Aggregate Asset Allocation Decisions, Working Paper, University of Oregon.

[30] Charness, G., and M. Dufwenberg, 2006, Promises and partnership, Econometrica 74, 1579-1601.

[31] Charness, G., L. Rigotti, and A. Rustichini, 2007, Individual behavior and group membership, American Economic Review 97, 1340-1352.

[32] Chen, J., H. Hong, and J. Kubik, 2011, Outsourcing mutual fund management: Firm boundaries, incentives and performance, Journal of Finance, forthcoming.

[33] Chen, J., H. Hong, M. Huang, and J. Kubik, 2004, Does fund size erode mutual fund performance? The role of liquidity and organization, American Economic

Review 94, 1276-1302.

[34] Chen, N., 1991, Financial Investment Opportunities and the Macroeconomy, Journal of Finance 46, 529-554.

[35] Chevalier, J., and G. Ellison, 1997, Risk taking by mutual funds as a response to incentives, Journal of Political Economy 105, 1167-1200.

[36] Chevalier, J., and G. Ellison, 1999a, Are some mutual fund managers better than others? Cross-sectional patterns in behavior and performance, Journal of Finance 54, 875-899.

[37] Chevalier, J., and G. Ellison, 1999b, Career concerns of mutual fund managers, Quarterly Journal of Economics 114, 389–432.

[38] Christoffersen, S., and S. Sarkissian, 2009, City size and fund performance, Journal of Financial Economics 92, 252-275.

[39] Cici, G., 2011, The relation of the disposition effect to mutual fund trades and performance, forthcoming in Journal of Financial and Quantitative Analysis.

[40] Cooper, D., and J. Kagel, 2004, Are two heads better than one? Team versus individual play in signaling games, American Economic Review 95, 477-509.

[41] Coval, J., and T. Moskowitz, 2001, The geography of investment: Informed trading and asset prices, Journal of Political Economy 109, 811-841.

[42] Dass, N., V. Nanda, and Q. Wang, 2011, Teams in fund management: Are they teams in name only? Working paper, Georgia Institute of Technology.

176

[43] Deuskar, P., J. Pollet, Z. Wang, and L. Zheng, 2011, The good or the bad? Which mutual fund managers join hedge funds? Review of Financial Studies 24, 3008- 3024.

[44] Duong, T., and F. Meschke, 2011, The rise and fall of portfolio pumping among U.S. mutual funds, Working paper, National University of Singapore.

[45] Edelen, R., and J. Warner, 2001, Aggregate Price Effects of Institutional Trading: A Study of Mutual Fund Flow and Market Returns, Journal of Financial

Economics 59, 195-220.

[46] Elton, E., M. Gruber, and C. Blake, 2001, A first look at the accuracy of the CRSP mutual fund database and a comparison of the CRSP and the Morningstar mutual fund database, Journal of Finance 56, 2415-2430.

[47] Elton, E., M. Gruber, S. Das, and M. Hlavka, 1993, Efficiency with costly information: a reinterpretation of evidence from managed portfolios, Review of

Financial Studies 6, 1-22.

[48] Epstein, L., and M. Schneider, 2008, Ambiguity, Informational Uncertainty and Asset Pricing, Journal of Finance 63, 197-228.

[49] Fama, E., and K. French, 1989, Business Conditions and Expected Returns on Stocks and Bonds, Journal of Financial Economics 25, 23-49.

[50] Ferson, W., and R. Schadt, 1996, Measuring fund strategy and performance in changing economic conditions, Journal of Finance 51, 425-461.

[51] Glaeser, E., 1999, Learning in cities, Journal of Urban Economics 46, 254-277.

[52] Gneezy, U., 2005, Deception: the role of consequences, American Economic

Review 95, 384-394.

[53] Goetzmann, W., J. Ingersoll, M. Spiegel, and I. Welch, 2007, Portfolio performance manipulation and manipulation-proof performance measures, Review of Financial Studies 20, 1503-1546.

[54] Greenwood R., and S. Nagel, 2009, Inexperienced investors and bubbles, Journal

of Financial Economics 93, 239-258.

[55] Grinblatt, M., and M. Keloharju, 2000, The Investment Behavior and Performance of Various Investor-Types: A Study of Finland's Unique Data Set, Journal of Financial Economics 55, 43-67.

[56] Grubb, D., and J. Symons, 1987, Bias in regressions with a lagged dependent variable, Econometric Theory 3, 71-386.

177

[57] Hamilton, B., J. Nickerson, and H. Owan, 2003, Team incentives and worker

heterogeneity: An empirical analysis of the impact of teams on productivity and participation, Journal of Political Economy 111, 465-497.

[58] Han, Y., T. Noe, and M. Rebello, 2008, Horses for courses: Fund managers and organizational structures, Working paper, Oxford University.

[59] He, J., L. Ng, and Q. Wang, 2004, Quarterly trading patterns of financial institutions, Journal of Business 77, 493-509.

[60] Hendricks, D., J. Patel, and R. Zeckhauser, 1993, Hot Hands in Mutual Funds: Short-Run Persistence of Relative Performance, Journal of Finance 48, 93-130.

[61] Hirshleifer, D, 2001, Investor Psychology and Asset Pricing, Journal of Finance 56, 1533-1597.

[62] Holmstrom, B., 1982, Moral hazard in teams, Bell Journal of Economics, 13, 324-340.

[63] Huang, J., C. Sialm, and H. Zhang, 2011, Risk shifting and mutual fund performance, Review of Financial Studies 24, 2575-2616.

[64] Ippolito, R., 1992, A consumer reaction to measures of poor quality: Evidence from the mutual fund industry, Journal of Law and Economics 35, 45-70.

[65] Jacob, B., and S. Levitt, 2003, Rotten apples: An investigation of the prevalence and predictors of teacher cheating, Quarterly Journal of Economics, 118, 843–77.

[66] Jacobs, J., 1969, The economy of cities, Vintage, New York.

[67] Jain, P., and J. Wu, 2000, Truth in Mutual Fund Advertising: Evidence on Future Performance and Fund Flows, Journal of Finance 55, 937-958.

[68] Janis, I., 1982, Groupthink: A psychological study of policy decisions and fiascoes, Houghton Mifflin Company, Boston.

[69] Jason, B., and S. Ludvigson, 1998, Does Consumer Confidence Forecast House- hold Expenditure? A Sentiment Index Horse Race, Federal Reserve Bank of New

York: Economic Policy Review 4, 59-78.

[70] Jehn, K., G. Northcraft, and M. Neale, 1999, Why differences make a difference: A field study of diversity, conflict and performance in workgroups, Administrative Science Quarterly 44, 741-763.

178

[71] Jensen, M., 1968, The performance of mutual funds in the period 1945-1964, Journal of Finance 23, 389-416.

[72] Kacperczyk, M., S. Van Nieuwerburgh, and L. Veldkamp, 2011, Rational Attention Allocation Over the Business Cycle." Working Paper, New York University.

[73] Kandel, E., and E. Lazear, 1992, Peer pressure and partnerships, Journal of

Political Economy 100, 801-817.

[74] Karagiannidis, I., 2010, Management team structure and mutual fund performance, Journal of International Financial Markets, Institutions & Money

20, 197-211.

[75] Kempf, A., and S. Ruenzi, 2007, Tournaments in mutual-fund families, Review of

Financial Studies 21, 1013-1036.

[76] Kerr, N., 1992, Group decision making at a multialternative task: Extremity, interfaction distance, pluralities and issue importance, Organizational Behavior

and Human Decision Processes 52, 64–95.

[77] Kocher, M. and M. Sutter, 2005, The decision maker matters: Individual versus group behavior in experimental beauty-contest games, Economic Journal 115, 200-223.

[78] Kostovetsky, L., and J. Warner, 2011, You’re fired! New evidence on portfolio manager turnover and performance, Working paper, University of Rochester.

[79] Lakonishok, J., A. Shleifer, R. Thaler, R. Vishny, 1991, Window dressing by pension fund managers, American Economic Review 81, 227-231.

[80] Laughlin, P., E. Hatch, J. Silver, and L. Boh, 2006, Groups perform better than the best individuals on letters-to-numbers problems: Effects of group size, Journal of Personality and Social Psychology 90, 644-651.

[81] Le Bon, G., 1896, The crowd: A study of the popular mind, Macmillan.

[82] Lemmon, M., and E. Portniaguina, 2006, Consumer Confidence and Asset Prices: Some Empirical Evidence, Review of Financial Studies 19, 1499-1529.

[83] Ludvigson, S., 2004, Consumer Confidence and Consumer Spending, Journal of

Economic Perspectives 18, 29-50.

[84] Ma, C., J. Moore, and S. Turnbull, 1988, Stopping agents from ‘cheating’, Journal of Economic Theory, 46, 355-372.

179

[85] Maddala, G., and A. Rao., 1973, Tests for serial correlation in regression models with lagged dependent variables and serially correlated errors, Econometrica 47, 761-774.

[86] Malmendier, U., and S. Nagel, 2011, Depression Babies: Do Macroeconomic Experiences Affect Risk-Taking?, Quarterly Journal of Economics 126, 373-416.

[87] Mas, A., and E. Moretti, 2009, Peers at Work, American Economic Review 99, 112-145.

[88] Massa, M., J. Reuter, and E. Zitzewitz, 2010, When should firms share credit with employees? Evidence from anonymously managed mutual funds, Journal of

Financial Economics 95, 400-424.

[89] Meier, I., and E. Schaumburg, 2006, Do funds window dress? Evidence for U.S. domestic equity mutual funds, Working paper, Northwestern University.

[90] Moscovici, S., and M. Zavalloni, 1969, The group as a polarizer of attitudes, Journal of Personality and Social Psychology 12, 125-135.

[91] Mueller, J., 2012, Why individuals in larger teams perform worse, Organizational

Behavior and Human Decision Processes 117, 111-124.

[92] Nalbantian, H., and A. Schotter, 1997, Productivity under group incentives: An experimental study, American Economic Review 87, 314-341.

[93] Ng, L., and Q. Wang, 2004, Institutional trading and the turn-of-the year effect, Journal of Financial Economics 74, 343-366.

[94] Nohel, T., Z. Wang, and L. Zheng, 2010, Side-by-side management of hedge funds and mutual funds, Review of Financial Studies 23, 2342-2373.

[95] Pastor, L., and R. Stambaugh, 2002, Investing in Equity Mutual Funds, Journal of

Financial Economics 63, 351-380.

[96] Patel, S., and S. Sarkissian, 2012, To group or not to group? Evidence from mutual funds, Working paper, McGill University.

[97] Petersen, M., 2009, Estimating Standard Errors in Finance Panel Data Sets: Comparing Approaches, Review of Financial Studies 22, 435-480.

[98] Pollet, J., and M. Wilson, 2008, How does size affect mutual fund behavior?” Journal of Finance, 63, 2941-69.

180

[99] Prather, L., and K. Middleton, 2002, Are N+1 heads better than one? The case of mutual fund managers, Journal of Economic Behavior and Organization 47, 103- 120.

[100] Rasmusen, E., 1987, Moral hazard in risk-averse teams, RAND Journal of

Economics 18, 428-435.

[101] Sah, R., and J. Stiglitz, 1986, The architecture of economic systems: Hierarchies and polyarchies, American Economic Review 76, 716–727.

[102] Sah, R., and J. Stiglitz, 1991, The quality of managers in centralized versus decentralized organizations, Quarterly Journal of Economics 106, 289–295.

[103] Santini, D., and J. Aber, 1998, Determinants of Net New Money Flows to the Equity Mutual Fund Industry, Journal of Economics and Business 50, 419-429.

[104] Shackle, G., 1942, A Theory of Investment-Decisions, Oxford Economic Papers 6, 77-94.

[105] Sharpe, W., 1981, Decentralized investment management, Journal of Finance 36, 217-234.

[106] Shleifer, A., 2004, Does competition destroy ethical behavior? American

Economic Review 94, 414-418.

[107] Sias, R. and L. Starks, 1997, Institutions and individuals at the turn-of-the-year, Journal of Finance 52, 1543-1562.

[108] Sirri, E., and P. Tufano, 1998, Costly search and mutual fund flows, Journal of

Finance 53, 1589-1622.

[109] Smith, R., 1971, Uncertainty, Information and Investment Decisions, Journal of

Finance 26, 67-82.

[110] Stein, J., 2002, Information production and capital allocation: Decentralized versus hierarchical firms, Journal of Finance 57, 1891-1921.

[111] Stoner, J., 1968, Risky and cautious shifts in group decisions, Journal of

Experimental Social Psychology 4, 442-459.

[112] Straughn, C., and B. Straughn, 1995, Lovejoy’s college guide, 23rd Edition, Macmillan, New York.

[113] Sunstein, C., 2002, The law of group polarization, Journal of Political

Philosophy 10, 175-195.

181

[114] Surowiecki, J., 2005, The wisdom of crowds, New York: Anchor Books.

[115] Sutter, M., 2009, Deception through telling the truth? Experimental evidence from individuals and teams, Economic Journal 119, 47-60.

[116] Thompson, L., 2003, Making the team, Upper Saddle Riverk, NJ: Pearson Education Inc.

[117] Thompson, S., 2011, Simple Formulas for Standard Errors that Cluster by Both Firm and Time, Journal of Financial Economics 99, 1-10.

[118] Van Knippenberg, D., and M. Schippers, 2007, Work group diversity, Annual

Review of Psychology, 58, 515-541.

[119] Veronesi, P., 2000, How Does Information Quality Affect Stock Returns?, Journal of Finance 55, 807-837.

[120] Wallach, M., and H. Kogan, 1965, The roles of information, discussion, and consensus in group risk taking, Journal of Experimental Social Psychology 1, 1-19.

[121] Warther, V., 1995, Aggregate Mutual Fund Flows and Security Returns, Journal

of Financial Economics 39, 209-235.

[122] Williams, K., and C. O’Reilly, 1998, Demography and diversity in organizations: A review of 40 years of research, Research in Organizational Behavior 20, 77– 140.

[123] Zhang, X., 2006, Information Uncertainty and Stock Returns, Journal of Finance 61, 105-137.

[124] Zheng, L., 1999, Is Money Smart? A Study of Mutual Fund Investors' Fund Selection Ability, Journal of Finance 54, 901-933.

[125] Zweig, J., 1997, Watch out for the year-end fund flimflam, Money Magazine, November, 130-133.