University of Southern California Dissertations and Theses

Investment behavior of mutual fund managers

(USC Thesis Other)

Investment behavior of mutual fund managers

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content INVESTMENT BEHAVIOR OF MUTUAL FUND MANAGERS
by
ROBIN Y. LEE
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(FINANCE)
May 2024
Copyright 2024 ROBIN Y. LEE

Acknowledgements
I am deeply indebted to my advisor, Kenneth Ahern, for his unwavering guidance, support, and
encouragement throughout my PhD journey. I also extend my sincere appreciation to David
Hirshleifer, Gerard Hoberg, Chris Parsons, and John Matsusaka for their consistent advice and
support. Additionally, I am thankful to my fellow PhD students at USC Marshall for their camaraderie and the shared experiences that enriched my academic journey.
The boundless love and support of my parents have been the cornerstone of my path, for
which I am forever grateful. Finally, I deeply thank my spouse, Taeho Lee, who has been the
greatest friend and companion in my life.
ii

Table of Contents
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
Chapter 1: The Effect of Social Interaction on Mutual Fund Managers: Prior Evidence . . 1
1.1 Face-to-face Social Interactions and Local
Informational Advantage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Collective Cognition of Mutual Fund Teams . . . . . . . . . . . . . . . . . . . . . . 8
Chapter 2: Face-to-face Social Interactions and Local Informational Advantage . . . . . . 13
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Identifying Face-to-face Communication Effect . . . . . . . . . . . . . . . . . . . . 20
2.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3.1 Mutual fund holdings and returns . . . . . . . . . . . . . . . . . . . . . . . 24
2.3.2 Location of fund managers . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3.3 COVID-19 lockdown information . . . . . . . . . . . . . . . . . . . . . . . 27
2.4 Impact of Lockdowns on Investment Decisions . . . . . . . . . . . . . . . . . . . . 30
2.4.1 Portfolio-level return . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.4.2 Investment timings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.5 Channel Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.5.1 Buy versus sell decisions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.5.2 Heterogeneity across stock informational environment . . . . . . . . . . . 37
2.5.3 Information flow within fund families . . . . . . . . . . . . . . . . . . . . . 39
2.5.4 Public information seeking . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.6 Impact of Lockdowns on Local Bias . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.7 Social Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
Chapter 3: Collective Cognition of Mutual Fund Teams . . . . . . . . . . . . . . . . . . . . 69
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
iii

3.3 Rank effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
3.3.1 Rank effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
3.3.2 Post-trade returns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.4 Rank effect in teams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.4.1 Team dummy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.4.2 Propensity score matching . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
3.5 Channel analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.5.1 Team member’s common holdings . . . . . . . . . . . . . . . . . . . . . . 88
3.5.2 Cognitive style heterogeneity . . . . . . . . . . . . . . . . . . . . . . . . . 90
3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
iv

List of Tables
2.1 Mutual fund characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.2 Change in footprint activities (%) relative to the 2019 average . . . . . . . . 59
2.3 Impact of lockdowns on local portfolio return . . . . . . . . . . . . . . . . . 60
2.4 Impact of lockdowns on stock-level investment returns . . . . . . . . . . . 61
2.5 Impact of lockdowns on buy/sell decision, trade size, and turnover ratio . 62
2.6 Speed of Information Diffusion (SID) within fund families . . . . . . . . . 63
2.7 Public information seeking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
2.8 Impact of lockdowns on local bias . . . . . . . . . . . . . . . . . . . . . . . . . 66
2.9 Social index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.1 Summary statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
3.2 Rank effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
3.3 Rank effect in teams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
3.4 Propensity score matching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
3.5 Rank effect on common holdings . . . . . . . . . . . . . . . . . . . . . . . . . 103
3.6 Paired t-tests of the rank effect of statistical team versus actual team . . . 104
v

List of Figures
2.1 Example of different investment performance on the same stock . . . . . . 49
2.2 Mutual fund manager location . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.3 Stay-at-home orders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.4 Footprint activities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.5 Impact of lockdowns on local portfolio return . . . . . . . . . . . . . . . . . 53
2.6 Impact of lockdowns on local stock-level return . . . . . . . . . . . . . . . . 54
2.7 Impact of lockdowns on the investment timings of buy orders . . . . . . . 55
2.8 Heterogeneity across stock informational environment . . . . . . . . . . . . 56
2.9 Local bias . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.10 Impact of lockdowns on local bias . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.1 Team management in the mutual fund industry . . . . . . . . . . . . . . . . 96
3.2 Post-trade returns relative to counterfactual . . . . . . . . . . . . . . . . . . 97
3.3 Propensity score matching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
vi

Abstract
This dissertation explores the role of social interactions in shaping fund managers’ investment
behaviors. Chapter 1 provides a review of the literature. Chapter 2 investigates the causal role
of face-to-face communication in generating local informational advantage. By exploiting variation in social interactions driven by COVID-19 lockdowns, I find that during lockdowns, mutual
fund managers’ performance on local stocks declined relative to non-local stocks. By establishing causality while ruling out alternative stories based on firm fundamentals and fund managers’
alternative information sources, the results suggest the importance of interpersonal interactions
for fund managers to acquire value-relevant information on local stocks. Chapter 3 explores the
social element within mutual fund management teams, providing evidence of the superior cognitive ability of mutual fund teams to mitigate a bias driven by constrained cognitive ability. By
leveraging data on the managerial history of mutual fund managers, I find a smaller rank effect in
team-managed funds relative to solo-managed funds. This reduction is particularly pronounced
for stocks that team members are more likely to discuss and in teams with cognitive style diversity. The results suggest a beneficial role of team decision-making in reducing reliance on
heuristics when making investment decisions.
vii

Chapter 1
The Effect of Social Interaction on Mutual Fund Managers:
Prior Evidence
The mutual fund industry has witnessed steady expansion over the past three decades, solidifying its position as one of the foremost financial intermediaries in the US. As such, mutual fund
managers represent a cohort of sophisticated professional investors within the financial market.
Consequently, much attention in financial literature has been devoted to comprehending their
performance delivery and decision-making processes.
Finance literature delineates two key concepts to understand the investment behavior of mutual fund managers. The first is the information advantage that they acquire through access to
superior information on specific segments of the portfolio, even though the overall portfolio may
not outperform the market on average. The second concept is the influence of behavioral biases,
wherein managers make systematic mistakes due to behavioral or psychological elements.
When exploring both aspects, an important factor to consider is the social dimension inherent
in the decision-making of professional investors. As social agents, investors do not operate in
isolation but interact with and learn from other agents (Hirshleifer 2020). Fund managers actively
1

participate in social interactions with various stakeholders to gather value-relevant information,
which may include visiting firms’ sites, socially interacting with corporate managers, employees,
managers, suppliers, customers, other peer investors, etc. Consequently, the role of these social
processes in shaping mutual fund managers’ investment behavior remains a topic of ongoing
discussion in the field of social finance literature.
This dissertation is structured as follows: Chapter 1 provides a literature review that lays
the groundwork for the essays in Chapters 2 and 3, which focus on two distinct types of social
interactions inherent in fund managers’ investment decision-making processes. In Chapter 2,
I examine the significance of face-to-face interactions arising in local areas, providing causal
evidence of the role of interpersonal social interactions on the generation of fund managers’
informational advantages on local firms. Chapter 3 explores the social dynamics within mutual
fund teams, addressing how team decision-making processes influence the behavioral bias arising
from fund managers’ cognitive constraints.
1.1 Face-to-face Social Interactions and Local
Informational Advantage
Since Jensen (1968), studies in mutual fund literature have shown little evidence that actively
managed mutual funds outperform passive benchmarks on average. However, these studies also
document that fund managers do have advantages in accessing superior information for a subset of portfolio firms, enabling them to demonstrate stock selection ability on such stocks and
generate superior returns.
2

One element of such an advantage documented in the literature is geography, which arises
from the observation that investors tend to favor investing in geographically proximate firms.
Coval and Moskowitz (1999) and Coval and Moskowitz (2001) were the first to document the bias
of mutual fund managers towards locally headquartered firms. By demonstrating that US fund
managers consistently earn substantial abnormal returns from investments in nearby firms compared to those located farther away, they provide evidence of asymmetric information based on
geographic distance and fund managers’ ability to exploit informational advantages in selecting
nearby stocks.
Recent studies examining the time periods in the 2000s continue to provide evidence of such
an advantage based on proximity. Despite improvements in the informational environment, including the implementation of Regulation Fair Disclosure and the widespread availability of the
internet, Baik, Kang, and Kim (2010) demonstrates that local institutional ownership still predicts
future stock returns, especially for firms characterized by high information asymmetry. Additionally, Bernile, Kumar, Sulaeman, and Wang (2019) shows that institutional investors still trade
local stocks disproportionately more often than nonlocal stocks, and that their local trades outperform nonlocal trades in the short term.
The importance of geographic distance in information acquisition and better monitoring is
documented across various participants as well. Proximate sell-side analysts demonstrate better forecasting abilities for investments in the US and Europe (Malloy 2005; Bae, Stulz, and Tan
2008). Banks provide more lending to local firms (Agarwal and Hauswald 2010), and corporate
managers monitor proximate plants more effectively (Giroud 2013). Additionally, venture capitalists’ local presence leads to increased innovation and a higher likelihood of successful exits
3

(Bernstein, Giroud, and Townsend 2016). Together, this extensive body of research underscores
the importance of geographic distance in information acquisition for financial agents.
However, geographic proximity may not only confer an informational advantage. Studies
document that individual investors’ preference for local stocks could be driven by familiarity
bias, leading to portfolio decisions based on irrational or non-information-based factors (Huberman 2001; Seasholes and Zhu 2010). Professional investors’ portfolio choices are also susceptible
to familiarity bias, as Pool, Stoffman, and Yonker (2012) documents that managers overweight
companies headquartered in the states where they grew up, which is not driven by information.
The dual nature of geography suggests that several factors underlying geographic proximity can
have differential effects on fund managers’ investment decision-making.
This raises the question of which aspects of geography play what roles in fund managers’
decision-making behavior. The inquiry stems from the diverse factors associated with proximity
to firms’ headquarters, which offer several possible explanations for common investment patterns
among investors in the same region. Investors often draw investment insights from local media
reports; Engelberg and Parsons (2011) documents a causal relationship between local newspaper
coverage of earnings announcements and subsequent local trading. Additionally, Kang, SticeLawrence, and Wong (2021) finds that investors acquire firsthand knowledge of the local economy
by observing timely physical indicators, such as the volume of cars in parking lots. Moreover,
considering that investors are not randomly distributed across different areas, individual characteristics—such as risk appetite, investment skills, and available resources—may contribute to
concerns of homophily, rather than solely attributing effects to proximity.
Another major factor related to geography that serves as an informational source is the social
network. This line of research focuses on information dissemination through person-to-person
4

relationships rather than through formal channels available to local agents. Several studies have
examined the role of social networks in influencing investment decisions. Cohen, Frazzini, and
Malloy (2008) provides evidence that connections between mutual fund managers and corporate board members through shared educational networks confer an informational advantage to
portfolio managers. Cohen, Frazzini, and Malloy (2010) highlights a school tie premium between
sell-side analysts and senior corporate officers. Hvide and Östberg (2015) documents coworkers
as a network through which stock information is disseminated. Ahern (2017) demonstrates the
flow of insider information through strong social ties, including family, friends, and geographic
proximity.
The literature goes further by providing evidence of a more direct channel for the dissemination of information: the effect of ’word-of-mouth’ communication, where information spreads
through social interactions based on the social network. Hong, Kubik, and Stein (2004) suggests
that regional social interaction is one of the underlying determinants of investors’ stock market
participation rates. Hong, Kubik, and Stein (2005) proposes that word-of-mouth communication influences mutual fund managers’ buying and selling decisions, under the assumption that
managers in the same city are more likely to interact directly with one another. Building upon
this, studies report consistent results while carefully addressing concerns regarding potential
unobserved characteristics. Brown, Ivković, Smith, and Weisbenner (2008) establishes the causal
link of the word-of-mouth channel using social capital from one’s birth region as an instrumental
variable, and Pool, Stoffman, and Yonker (2015) employs granular location information to demonstrate correlated and profitable trading activities among neighboring mutual fund managers.
In this course of discussion on the role of social interaction, what is still lacking is an examination of the various forms it can take, particularly given advancements in communication
5

technology. Even if individuals have established social networks through physical meetings in
the local area, they may still rely on electronic means such as messages, phone calls, emails, and
Zoom meetings. Furthermore, the ease of travel has diminished the distance between economic
agents, thereby altering the concept of proximity, as shown in studies that exploit the introduction of direct airline flights and high-speed railways (Ellis, Madureira, and Underwood 2020; Da,
Gurun, Li, and Warachka 2021; Chen, Qu, Shen, Wang, and Xu 2022). Adding to the prior studies exploring the role of private interpersonal meetings in information acquisition for hedge fund
managers (Solomon and Soltes 2015) and sell-side analysts (Choy and Hope 2021), the question of
the role of face-to-face meetings in shaping mutual fund managers’ investment decision-making
regarding local firms remains to be explored.
The role of face-to-face meetings is interesting to explore given their distinctive features.
Face-to-face communication fosters the development and sustenance of trust, enabling the sharing of rich information, which is particularly important in an environment where information is
imperfect and rapidly changing (Storper and Venables 2004). Moreover, the unique non-codifiable
traits of face-to-face communication make them not fully substitutable by other forms of social
interactions. In finance literature, several studies have shown that non-verbal cues, including
body language and facial expressions (Hu and Ma 2021; Peng, Teoh, Wang, and Yan 2021), vocal
tones (Bartov, Faurel, and Mohanram 2018), and response time (Frydman and Krajbich 2022), play
a significant role in information transmission among economic agents.
Another reason to consider that social interactions arising in face-to-face settings could substantially impact fund managers’ decision-making is the potential for selective communication
among local agents: favorable cues often dominate social interactions. As documented by Hirshleifer (2020), social transmission bias alters the signals or ideas during social interactions. Han,
6

Hirshleifer, and Walden (2022) documents selective communication in the stock market, where
profitable investment ideas are more likely to be transmitted and recounted. This idea aligns with
findings in the literature on communication about consumer products, where positive word-ofmouth discussions prevail over negative ones, reflecting the concept of impression management
(Leary and Kowalski 1990; East, Hammond, and Wright 2007; Wojnicki and Godes 2017). Given
the persistence of this phenomenon in one-on-one meetings, the presence of such selectivity
in information flow from portfolio firms to fund managers could be beneficial for mutual fund
managers in selecting which local stocks to add to their portfolios to produce superior returns.
Building upon this line of thought, the essay in Chapter 2 delves into the role of face-to-face
meetings occurring in local physical settings, identifying them as a crucial source of information advantage that cannot be fully replaced by other forms of social interaction that mutual
fund managers engage in. Utilizing the Covid lockdowns in early 2020 as a shock that created
exogenous variation in the extent of face-to-face social interaction among local agents, I establish a causal relationship between local in-person interactions and mutual fund managers’ local
informational advantage. By carefully isolating the effect from that of firm fundamentals or alternative information sources, I provide evidence that even in an era of advanced communication
technologies, the comprehensive sharing of stock information cannot be fully sustained in the
absence of ongoing face-to-face social interactions.
7

1.2 Collective Cognition of Mutual Fund Teams
The second aspect that social interaction could influence professional investors’ decision-making
is their susceptibility to behavioral biases. Prior research has documented various behavioral biases exhibited by professional money managers. They tend to purchase overvalued stocks (Edelen, Ince, and Kadlec 2016) and are prone to the disposition effect, wherein they tend to realize
gains more readily than losses (Cici 2012), which is exacerbated when they are distracted (Lu,
Ray, and Teo 2016). Moreover, Puetz and Ruenzi (2011) finds that they exhibit overconfidence,
trading excessively following good performance. Professional investors also face the problem of
limited attention, trading attention-grabbing stocks covered by mass media (Fang, Peress, and
Zheng 2014; Ben-Rephael, Da, and Israelsen 2017), extreme winning and losing positions (Hartzmark 2015), and becoming distracted with too many positions in a portfolio having earnings
announcements (Schmidt 2019).
However, the discussion of fund managers’ behavioral biases lacks an examination of the
entities involved in decision-making and the social processes underlying such decisions. This
is particularly relevant given the evolving structure of fund management over the last decades.
The industry has witnessed a shift towards team management as a dominant decision-making
unit: while team management accounted for less than 20% before 1990, it has steadily increased
to nearly 80% by 2020. If fund investment decisions and outcomes reflect the decision-making
processes at the group level, the impact of social interaction among team members needs to be
considered when exploring the behavioral and psychological biases that drive their mistakes.
The question of whether team decision-making can mitigate fund managers’ behavioral biases
remains open, as there is no consensus in the literature regarding whether teams can lead to
8

superior decision-making. Studies document several reasons that may contribute to less optimal
decision-making in teams, including risky shifts (Kogan and Wallach 1967; Stoner 1968), group
polarization (Moscovici and Zavalloni 1969), and groupthink, wherein team members may accept
predominant but suboptimal decisions due to social pressure for unity (Janis 1982; Bénabou 2013).
Empirical studies that do not find superior performance in team-managed funds compared to
those managed by solo fund managers support this line of thoughts (Prather and Middleton 2002;
Chen, Hong, Huang, and Kubik 2004; Bliss, Potter, and Schwarz 2008; Adams, Nishikawa, and
Rao 2018; Massa, Reuter, and Zitzewitz 2010; Bär, Niessen-Ruenzi, and Ruenzi 2007).
On the other hand, in line with the findings of Patel and Sarkissian (2017) that team-managed
funds outperform single-managed funds across various performance metrics with comprehensive data, another line of studies documents that the diversity of mutual fund team members
can add value in asset management. Harvey, Liu, Tan, and Zhu (2021) finds that team members
with heterogeneous skills and experiences can bring in fresh ideas to exploit more varied investment opportunities, and are less susceptible to capacity constraints. Studies also document the
beneficial role of diversity in other dimensions, such as demographic or socioeconomic characteristics including gender, ethnicity, education, ideology, etc., (Bär, Niessen-Ruenzi, and Ruenzi
2007; Evans, Prado, Rizzo, and Zambrana 2020; Lu, Naik, and Teo 2024).
As prior studies find that teams of professional managers can pool resources and information
to improve decision-making, a related question is whether they could also pool cognitive abilities
to mitigate biases arising from their constrained cognitive resources. Investing decisions entail a
sequence of choices that require cognitive resources, which is scarce (Kahneman 1973), including
attention, memory, and processing power. Consequently, fund managers often make cognitive
errors based on heuristic simplifications, whereby the brain shortcuts complicated analyses. This
9

phenomenon, being one of the three elements explaining psychological biases studied in behavioral finance (Hirshleifer 2015), suggests that errors based on heuristic simplification may be
mitigated with enhanced cognitive abilities if fund managers can pool such resources in a team.
The idea of collective cognition in psychology literature, defined as the group process involved
in the acquisition, storage, transmission, manipulation, and use of information in a group (Von
Cranach, Ochsenbein, and Valach 1986; Wegner 1987; Hinsz, Tindale, and Vollrath 1997), suggests
the possibility that fund manager teams are more cognitively sophisticated than individuals. In
line with this idea, the interactions in team decision-making could engage the deliberate and
slow System 2 in the dual process theory of cognition (Sloman 1996; Stanovich and West 2000;
Kahneman and Frederick 2004; Kahneman 2011), resulting in enhanced cognitive performance in
a team of fund managers.
Moreover, the concept of cognitive style—the way individuals encode, process, and communicate information (Ausburn and Ausburn 1978; Kozhevnikov, Kosslyn, and Shephard 2005)—suggests heterogeneity in individuals’ cognitive styles. The finding that diversity of cognitive styles
positively affects the collective cognition of teams (Aggarwal, Woolley, Chabris, and Malone 2019)
further suggests that biases arising from cognitive constraints may be mitigated in fund manager
teams with diverse cognitive styles, which would add to prior studies documenting the beneficial
role of diversity among mutual fund team members in other dimensions.
If team members’ collective cognition can alleviate attention constraints, prior studies suggest
that such benefits would manifest more prominently in the selling behavior of fund managers,
which is known to be more susceptible to behavioral biases. Studies document the use of heuristics when investors make selling decisions compared to buying decisions. These decisions are
often less disciplined, as investors tend to allocate less attention to them (Grosshans, Langnickel,
10

and Zeisberger 2020; Tosun, Jin, Taffler, and Eshraghi 2022). It is suggested that poor selling
decisions stem from investors’ asymmetric allocation of cognitive resources rather than a lack
of fundamental skill in making optimal selling decisions (Akepanidtaworn, Di Mascio, Imas, and
Schmidt 2021), which provides a possibility of improvement through enhanced cognitive ability
within teams.
Among several documented behavioral biases of mutual fund managers, the rank effect
(Hartzmark 2015) offers a useful phenomenon to explore whether team members could pool their
limited cognitive resources to reduce errors in making selling decisions. The rank effect refers to
a propensity where investors sell extreme winning and losing positions while ignoring positions
in the middle, based on the stocks’ return from the purchase price. This effect is attributed to
the use of heuristic simplification in intuitive judgments. Given that extreme ranks are salient
(Diecidue and Wakker 2001), managers subconsciously include extreme-ranked positions as candidates for selling. Relative to a few recent studies documenting the role of mutual fund teams
in the effect of biases likely to be caused by behavioral preferences (Fedyk, Patel, and Sarkissian
2023; Barahona, Cassella, and Jansen 2022; Dorn and Yadav 2022), focusing on a bias likely to be
caused by cognitive constraints provides a useful framework for exploring the beneficial role of
enhanced collective cognition in mutual fund teams.
Building on this, Chapter 3 provides evidence of the superior cognitive ability of mutual fund
teams to mitigate the rank effect. By exploiting data on the managerial history of mutual fund
managers, I find a smaller rank effect in team-managed funds relative to solo-managed funds
after controlling for fund family, fund, and fund manager characteristics. Furthermore, the reduction is more pronounced for stocks that team members are more likely to discuss and in teams
11

with cognitive style diversity. This provides evidence of the beneficial impact of collective cognition, which involves social interaction within teams, in reducing reliance on heuristics in asset
management.
12

Chapter 2
Face-to-face Social Interactions and Local Informational
Advantage
2.1 Introduction
It is well established that institutional investors prefer to invest in geographically proximate firms.
Although the preference may result from behavioral biases such as familiarity, prior studies document the existence of local informational advantages possessed by professional investors.∗ Despite the ease of communication afforded by technological progress and strict disclosure rules, the
continued local preference suggests that face-to-face contact is important for the transmission of
value-relevant information.
In addition to the nature of in-person communication that leaves no paper trails, face-to-face
interactions have benefits that cannot be easily achieved by other means of communication. The
cues that arise in in-person interactions, including body language, facial expressions, and vocal tones, facilitate the transfer of tacit and non-codified knowledge. Several studies document
∗See Huberman (2001), Seasholes and Zhu (2010), and Pool, Stoffman, and Yonker (2012) for studies on familiarity
bias, and Coval and Moskowitz (1999), Coval and Moskowitz (2001), Baik, Kang, and Kim (2010), and Bernile, Kumar,
and Sulaeman (2015) for studies on the local informational advantage of professional investors.
13

that such non-verbal cues account for a large part of information transmission among managers,
investors, analysts, and other stakeholders.† Moreover, face-to-face meetings foster the development and sustenance of trust and strong social relationships (Cummings, Butler, and Kraut 2002;
Urry 2003; Storper and Venables 2004), which allows for the sharing of rich information.
Previous research on information sharing among professional investors document the importance of proximity by using geographic distance to identify the likelihood of social interactions
(Hong, Kubik, and Stein 2005; Pool, Stoffman, and Yonker 2015). However, these studies rely on
cross-sectional variation in the distance between agents, which may correlate with time-invariant
omitted variables such as risk appetite, investment skills, and resources. Also, cross-sectional
variation captures multiple ways in which proximity matters. For example, nearby investors are
exposed to common information sources including local media (Engelberg and Parsons 2011) and
the observation of local economic areas (Kang, Stice-Lawrence, and Wong 2021).
More importantly, the cross-sectional variation does not disentangle the role of face-to-face
communication from that of electronic communication not involving face-to-face meetings, as
local agents who have established social connections through past face-to-face meetings can
communicate using technology. Therefore, proximity reflects the intensity of social interactions
among nearby agents through various means, and cross-sectional variation alone does not identify the causal role of continual face-to-face communications from that of other forms of social
interactions.
In this paper, I address these challenges by exploiting the interruption in face-to-face meetings
initiated by the unexpected outbreak of COVID-19 in early 2020. To curb the spread of disease,
†See Roberts, Sanderson, Barker, and Hendry (2006), Bushee, Jung, and Miller (2011), Mayew and Venkatachalam
(2012), and Peng, Teoh, Wang, and Yan (2021) for related works.
14

US county and state governments implemented stay-at-home orders. The lockdowns are an attractive setting for this study because they gave rise to cross-sectional and time-series variations
in fund managers’ in-person social interactions, allowing for a Difference-in-Differences (DiD)
research design.
Using lockdown orders might be subject to measurement error if people violate orders or if
they refrain from physical meetings before the orders are implemented. To address this problem, I
use SafeGraph footprint data and the number of Covid cases to proxy for face-to-face interactions.
Safegraph tracks the number of visits to 3.6 million commercial points-of-interests (POIs) in the
US using 45 million anonymous mobile devices. Given that the median footprint percentage
change in March 2020 is −25%, I assign funds and firms to the treatment group if the footprint
activities of their location decrease by more than 30% relative to the 2019 average.
To identify the role of face-to-face contact, it is crucial to identify the correct location of
the fund managers making the day-to-day investment decisions, as they are often distant from
the headquarters of fund management companies. I start by hand-collecting the residential zip
codes of fund managers from the LexisNexis Public Records Database, which are then crosschecked with their LinkedIn profiles. I classify any stock headquartered within 100 miles of a
fund manager’s zip code as local to the fund. I set the main sample period from January 2019 to
June 2020 to focus on the early periods of the pandemic during which fund managers were most
likely to refrain from physical interactions.
This paper exploits the natural experiment to investigate the empirical prediction that disrupted information flow from local firms to fund managers during lockdowns would cause fund
managers’ trading decisions on local stocks to deteriorate. To test the idea, I begin by providing
suggestive evidence with portfolio-level returns. I find that the benchmark excess return and
15

characteristic-adjusted (DGTW) return (Daniel, Grinblatt, Titman, and Wermers 1997) of a local
portfolio is lower than that of a distant portfolio by 0.4-0.7 percentage points on average when
a fund manager resides in a locked-down area. The changes are non-trivial as the difference
between local and distant portfolios became about three times bigger than in the pre-lockdown
period.
However, the results are naïve in that they do not take into account different stock compositions of local and distant portfolios of fund managers in different areas. If the firms located in
locked-down areas experience worse local economic conditions, their stock values would decline,
which would adversely impact the local portfolio return of fund managers residing in lockeddown areas. If this were the case, finding a negative treatment effect could be a spurious result of
the diminished stock returns of the firms in locked-down areas, not the result of fund managers’
poor trading decisions due to curtailed face-to-face interactions.
To address the concern, I run stock-level tests and compare two fund managers’ investment
timings on the same stock before and after lockdowns, where one manager is local and another is
non-local to the firm. To measure investment timings, I use the percentage change in the dollar
value of holdings, adjusting for the initiation of new positions, as the dependent variable in the
DiD regression. Stock×Post-lockdown fixed effects are included to control for all time-invariant
stock traits before and after a firm’s local area is locked down. As the fixed effects absorb the
effect of the underlying stock returns over a short time span before and after lockdowns, the
regression isolates the timing decision of managers while holding fixed all other aspects of the
investment. Specifically, fund managers who buy before the stock price increases and sell before
the price decreases would display superior stock-level investment performance.
16

I find that local fund managers’ three-month investment performance is lower by 0.04 percentage points during lockdowns compared to distant managers investing in the same stock,
compared to the pre-lockdown period. This translate to an economic loss of $0.4 million per
portfolio on average on an annual basis. The results imply that fund managers’ poor investment
performance on local stocks during lockdowns is driven by their deteriorated investment timings
on local stocks and not by the changes in firm fundamentals during lockdowns.
Next, I study the channels through which face-to-face interactions matter for local informational advantages. By regressing the (signed) next-period stock returns on a triple interaction
term, Local×Post-lockdown×Buy, I find that fund managers’ buy decisions were more significantly negatively impacted than sell decisions. At the same time, I do not find significant changes
in the size of trades and turnover ratios on local versus distant positions, which indicate that the
deteriorated timing on buy-orders is not driven by changes in fund managers’ aggressiveness or
activeness in making trading decisions. The results suggest that information transmitted in faceto-face settings is more positive in nature, and fund managers use this advantage when adding
local positions to their portfolios. The results are consistent with mutual funds’ short-sale constraints and the nature of face-to-face meetings, in which impression management facilitates the
transmission of positive information.‡
If fund managers’ deteriorated trading decisions during lockdowns are due to the loss of access to the information profitable for buy-orders, the results would be pronounced for stocks
that are informationally sensitive. I find that the worse investment timings on buy-orders arise
bigger for the stocks with big pre-lockdown trade sizes, which proxy for fund managers’ access
‡See Leary and Kowalski (1990), East, Hammond, and Wright (2007), Hirshleifer (2020), and Han, Hirshleifer,
and Walden (2022) for the discussion on impression management and its application in the stock market.
17

to superior information before lockdowns. Also, the effects are pronounced for the stocks with
high idiosyncratic risks, less publicly available information, high transaction costs, and large dispersion of belief, which are measured by idiosyncratic volatility, firm size, Amihud illiquidity,
and analyst forecast dispersion. The results suggest that face-to-face communication helps fund
managers obtain information on local firms with less transparent informational environments.
Next, I consider the possibility that the results are driven by a decrease in information from
alternative sources during lockdowns, the information spread within fund families and fund managers’ reliance on public information. Although I find that the information diffusion within fund
families slowed during lockdowns, there are no significant differences in the results on investment
timings across funds with different pre-lockdown speeds of information diffusion. Furthermore,
managers’ reliance on analysts’ recommendations did not decrease during lockdowns, which together help rule out the possibility that a decrease in information from other sources drives fund
managers’ poor local performance during lockdowns.
Additionally, as fund managers’ local overweighting does not change with lockdowns on average, I examine how the dispersion of local overweighting across funds changes with lockdowns.
To test whether fund managers’ local biases converge during lockdowns, I compute each fund’s
monthly local bias following Coval and Moskowitz (2001) and categorize funds into three groups
based on their local bias during 2019. I find that funds with the least pre-pandemic local bias
increased their local bias by 17-19%, while the funds with the highest pre-pandemic local bias
decreased their local bias by 4-5% during lockdowns. The convergence of fund managers’ preferences for local stocks during lockdowns corroborates the main result that local face-to-face
interactions play an important role in explaining fund managers’ investment decisions in proximate firms.
18

Finally, I provide auxiliary evidence that the results on investment performance and local bias
are more pronounced for funds in regions with stronger social ties, characterized by a higher
number of venues where face-to-face interactions are likely to occur with greater intensity. This
supports the idea that interpersonal social interaction is an important factor in determining mutual fund managers’ trading behavior on local stocks.
The central contribution of this paper is to present causal evidence on the role of face-to-face
interactions in generating local informational advantage. While previous research on professional investors’ local preferences and their social interactions exploits cross-sectional variation
in distance between agents to document the importance of proximity in information transmission,§
the natural experiment with the time-series component that I exploit allows the isolation of
the role of continual face-to-face interactions from that of social connections and communication
in other forms.
Moreover, while previous studies have primarily focused on providing evidence of performance outcomes, this paper explores the channels through which face-to-face information sharing matters for gaining an informational advantage. By exploring fund managers’ trading behavior and the cross-sectional heterogeneity across stocks with different information environments,
I provide evidence on the channels in which the information is advantageously used by fund
managers.
§See Coval and Moskowitz (1999), Coval and Moskowitz (2001), Malloy (2005), Bae, Stulz, and Tan (2008), Baik,
Kang, and Kim (2010), Engelberg and Parsons (2011), Bernile, Kumar, and Sulaeman (2015), Bernile, Kumar, Sulaeman, and Wang (2019), and Sialm, Sun, and Zheng (2020) for professional investors’ local bias, and Hong, Kubik, and
Stein (2005), Brown, Ivković, Smith, and Weisbenner (2008), Cohen, Frazzini, and Malloy (2008), Han and Yang (2013),
Pool, Stoffman, and Yonker (2015), Ahern (2017), Crawford, Gray, and Kern (2017), and Han, Hirshleifer, and Walden
(2018) for the role of social interactions in stock information transmission. Relatedly, see Giroud (2013), Bernstein,
Giroud, and Townsend (2016), Ellis, Madureira, and Underwood (2020), Da, Gurun, Li, and Warachka (2021), Choy
and Hope (2021), and Chen, Qu, Shen, Wang, and Xu (2022) for reduced information asymmetry through traveling.
19

A concurrent paper by Bai and Massa (2022) also studies the effect of Covid shutdowns on
mutual fund managers’ investment performance on proximate stocks using the headquarters locations of fund management companies and the fund-level average holding distance, concluding
that soft information cannot be substituted by hard information. My approach better identifies
the role of face-to-face as I use fund managers’ precise locations to identify local stocks, which is
critical as face-to-face interactions are likely to occur in very proximate areas. Moreover, my research design at the stock level provides evidence that the results are not driven by the change in
underlying stock returns due to Covid lockdowns, and the channel analyses provide a rich understanding of how the information shared face-to-face matters in generating a local informational
advantage for mutual fund managers.
2.2 Identifying Face-to-face Communication Effect
To illustrate the main challenge in identifying the role of face-to-face interactions in fund managers’ decision-making, consider mutual fund managers based in Los Angeles. Before the outbreak of COVID-19 in early 2020, they freely had in-person social interactions with local CEOs,
employees, customers, and other investors, who may share value-relevant information on local
firms. They were also able to visit nearby firms and directly observe operations. However, when
Los Angeles was locked down in March 2020 with the implementation of a stay-at-home order,
they were no longer able to engage in face-to-face communication in the area.
As a result of the lockdown, the LA-based fund managers lost one critical source of information on local firms. On the other hand, their information environment on non-local firms
remained the same as before because their primary informational channel on distant firms would
20

be electronic. In the same sense, non-LA based fund managers’ informational environment on
the firms headquartered in LA remained the same as before the lockdown.
Fund managers at different geographical locations experienced the lockdown with slightly
different timings. For example, fund managers based in LA were locked down starting March
2020, while those based in Houston were locked down starting April 2020. In addition to the
cross-sectional variation based on distance, such time-series variation in the amount of local
face-to-face meetings that lockdowns generate allows me to employ a Difference-in-Differences
(DiD) method.
As limited in-person activities disrupt the flow of information from local firms but not from
distant firms, the lockdowns would affect fund managers’ investment performance on local stocks
but not on distant stocks. Therefore, I divide each fund’s monthly portfolio into two portions,
local and non-local, and compare the change in the performance of the two portions by running
DiD regressions. The analysis compares the changes in fund managers’ investment performances
from the pre-lockdown to the post-lockdown period between local and distant investments.
Although the lockdowns are a useful setting for this study, several concerns arise when establishing a causal relationship between face-to-face communication and investment performance.
One concern is omitted variable bias, as there may be other factors affecting both fund managers’
social interactions and local investment performance. Ideally, treatment should be random in
that funds and stocks are randomly allocated into the locked-down and non-locked-down groups.
Since this is not the case, I include fund fixed effects to control for unobservable time-invariant
traits of fund managers, which rules out alternative explanations such as fund managers’ political
orientation or risk aversion, and allows the investigation of within-fund time-series variations.
21

Another concern with using COVID-19 as a quasi-experiment is that the pandemic is a macrolevel shock that affects the stock return of the firms they invest in. To address the issue, I include
year-month fixed effects to capture overall macro changes in stock returns.
A remaining concern is that portfolio-level tests do not take into account different stock compositions of local and distant portfolios of fund managers in different areas. If the firms located in
locked-down areas experience worse local economic conditions, their stock values would decline,
which may partly explain fund managers’ poor investment performance on local stocks. If this
were the case, finding a negative treatment effect would be a spurious result of the deteriorated
fundamentals and diminished stock returns of the firms after lockdowns, not the result of fund
managers’ poor trading decisions due to curtailed face-to-face interactions.
To address the concern, I run stock-level tests and compare two fund managers’ investment
timings on the same stock before and after lockdowns, where one manager is local and another is
non-local to the firm. Figure 2.1 illustrates how different investment decisions can lead to varying
investment returns for the same stock. The blue solid line represents the three-month return of
a stock. The green long-dashed and red short-dashed lines depict the three-month returns of
two funds that invested in the stock. These returns are calculated as the cumulative monthly
percentage changes in the dollar value of holdings, assuming that new positions are bought or
sold at the previous month’s end price. In this example, both funds initially held the same position
in the stock but executed additional trades at different times, resulting in different time series of
returns. Fund 1 doubled its position during January 2020, after which the stock had a positive
return at the end of the month. Consequently, Fund 1 achieved a higher return than the stock
itself due to well-timed trades, while Fund 2, which did not adjust its position, had the same
return as the stock. However, when Fund 1 sold its position during April 2020 after which the
22

stock’s return was positive, Fund 1’s return becomes lower than that of the stock. In contrast,
when Fund 2 doubled its position during May 2020, the stock experienced significantly positive
returns afterward. Consequently, Fund 2 achieved a higher return than both the stock and Fund
1 due to good investment timing.
In the main analyses, I compute each fund’s three-month investment returns on each stock,
obtained by cumulating each fund’s monthly returns of the dollar value invested on each stock,
adjusting for the initiation of the new positions, and run DiD regressions with the return as the
dependent variable. With the panel data in which the same stock is invested by the same set of
managers, some of them having their face-to-face interaction interrupted, I am able to include
Stock×Post-lockdown fixed effects to control for all time-invariant stock traits, including the
underlying stock returns over a short time span before and after lockdowns. With the fixed effects,
the regression isolates the timing decision of managers while holding fixed all other aspects of
the investment. Specifically, fund managers with well-timed trades, buying at low prices and
selling at high prices, would display superior stock-level investment performance.
The fixed effects solve several challenges. For example, while it is true that the stock price of
tech firms surged during the sample period, which may inflate the local investment performance
of fund managers located in the Bay Area, this is not a concern because the non-local investors
investing in the tech stocks are exposed to the same stock price changes. Also, the tech stocks
being non-local to most of the fund managers is irrelevant to my results because whether a fund
manager is local or non-local to a specific firm remains the same before and after lockdowns. For
the same reason, worries coming from politics affecting lockdown policies and thus the performance of local firms can be dismissed.
23

2.3 Data
This study requires data of three types: (1) Mutual fund holdings and returns (2) Location of
mutual fund managers (3) COVID-19 lockdown information.
2.3.1 Mutual fund holdings and returns
The first primary source of data in this paper is the CRSP survival-bias-free Mutual Fund Holdings
database, which is the most up-to-date database on mutual fund holdings information. Sample
funds for the main analyses are limited to actively managed US equity funds.
To select qualified funds, I filter Morningstar style categories following Pool, Stoffman, and
Yonker (2015). Only the funds with a Morningstar category in the 3-by-3 Size (Large, Mid, and
Small) and Value (Blend, Growth, and Value) grid are chosen. Funds with fewer than 20 holdings
or more than 500 holdings are removed as funds with more than 500 holdings could be index
funds. I consider only distinct portfolios by removing duplicated funds within the same fund
family with identical portfolios but with different share classes.
I further filter out funds using CRSP Lipper objective codes to remove non-equity funds, index
funds, ETFs, global and region funds, balanced funds, and sector funds. Funds that do not invest
primarily in equity, holding less than 50% in common or preferred stocks, are removed. Following Coval and Stafford (2007), funds whose total net assets double or halve from one quarter to
another are excluded. Funds that manage less than $1 million are removed. To avoid distance
outlier effects, funds with managers located in Alaska, Hawaii, Puerto Rico, the Virgin Islands, or
foreign countries are dropped. Finally, I exclude funds if the whole management team has been
24

replaced by another team, or if funds do not have any local stock holdings. After all of these
screens, 1,037 funds remain.
While mutual funds are required to report their quarterly holdings to SEC, some funds voluntarily disclose monthly holdings. About 45% of the sample funds have monthly holdings, while
another 45% have quarterly holdings. The remaining 10% have a mixture of monthly and quarterly
holdings. To obtain monthly holdings information, I forward-fill missing holding information at
the monthly frequency. The forward filling allows me to capture only the partial impact of lockdowns on the change in portfolio holdings, but it does not significantly bias my result as the main
sample period covers at least one quarter after lockdowns.
To compute a fund’s return on its local and non-local positions, I retrieve stock returns from
CRSP, and identify a fund’s benchmark index based on Morningstar investment style. I find
other firm-level information including the zip code of firm headquarters from Compustat. Only
the firms headquartered in the US with share code 10 or 11 remain in the sample. This leaves
3,524 stocks in the sample. To minimize the influence of outliers, all continuous variables are
winsorized at the 1% level.
2.3.2 Location of fund managers
The second major source of data in this paper is the LexisNexis Public Records Database (LNPRD).
The database contains extensive demographic information, including a history of all addresses
associated with each person, drawing on public records from county tax assessor records, state
motor vehicle registrations, reports from credit agencies, court filings, and post office records
among other sources. The database has been used in finance research to identify social networks
25

(Ahern 2017), hometown states (Pool, Stoffman, and Yonker 2012; Yonker 2017; Jiang, Qian, and
Yonker 2019), and residential addresses of investors (Pool, Stoffman, and Yonker 2015). Using
the database, I identify the exact location of the fund managers making day-to-day investment
decisions.
I first obtain information on each fund manager from MorningStar, which reports their full
name, start and end dates with the fund, educational background, and employment history. With
this information, I find the zip codes of fund managers’ home addresses from LNPRD. For the
funds that are team-managed by more than one fund manager, I use the zip code of the lead
manager if all managers reside in the same city. If not, I keep the zip codes of every manager in
different cities. As a result, 12% of the sample funds have more than one identified location. The
location of the managers is cross-checked with city information on their LinkedIn profiles. For
managers with more than one home address, I choose the closest to the city identified in their
LinkedIn profiles.
I identify fund managers’ residential zip codes for 69% of sample funds. For the fund managers
whom I cannot identify on LNPRD, I use their office address zip codes located in the city identified
in LinkedIn, which often differ from the headquarters location of fund management companies.
I identify local stocks by calculating the distance between a fund manager and firm headquarters
using the latitude and longitude of the centroid of zip codes. A stock is defined as local if the
distance is shorter than 100 miles. If a fund is managed by several fund managers in different
areas, firms within 100 miles of any of the locations are defined as local to the fund.
Table 2.1 reports the characteristics of sample funds separately for the pre-pandemic period
(Jan 2019 - Feb 2020) in Panel (a) and the post-pandemic period (Mar 2020 - Jun 2020) in Panel
(b). The median fund is managed by two fund managers residing in the same city, and the most
26

extreme fund has 24 managers in seven different cities. For an average fund, about 12% of the
stock holdings are within 100 miles.
Figure 2.2 plots the location of fund lead managers across the US at the state and county
levels. They are located in 470 different zip codes across 162 counties in 39 states and the District
of Columbia. The circles in Panel (a) represent the number of fund managers in each state. The
sizes of New York and Massachusetts are scaled down by half, which are the most common states,
each accounting for 18% of sample funds. California is the next common state with 10%, followed
by Illinois 7%, Texas 5%, Pennsylvania 5%, and Ohio 3%. There are no managers in nine states
in the sample, which are Nevada, Idaho, Wyoming, South Dakota, Arkansas, Mississippi, South
Carolina, West Virginia, and Rhode Island.
2.3.3 COVID-19 lockdown information
I collect three types of information to measure changes in face-to-face communication brought
about by COVID-19 induced lockdowns: stay-at-home orders, footprint activities, and the number of Covid cases.
The first data is stay-at-home orders which required people to remain at their places of residence. The stay-at-home orders issued by the US state governments are retrieved from COVID-19
US state policy database (CUSP) and cross-checked with a New York Times article that tracked
the orders across the US.¶ To adjust for the areas without statewide orders or local orders that
preceded statewide orders, I adjust the orders at the county level using data from the National
Association of Counties–County Explorer (NACo).
¶
“See Which States and Cities Have Told Residents to Stay at Home”
https://www.nytimes.com/interactive/2020/us/coronavirus-stay-at-home-order.html
27

Figure 2.3 shows the adoption of stay-at-home orders across the US in March and April of
2020. Red indicates the state with orders, blue with no orders, and grey indicates the states with no
fund managers in the sample. Most states issued lockdown orders except for six states: Arkansas,
Iowa, Nebraska, North Dakota, South Dakota, and Wyoming. Some counties in Florida, Indiana,
Pennsylvania, Utah, and Wyoming had different orders from the state-level orders. In March
2020, lockdown orders were implemented in 27 states out of 40 states in which fund managers
are located. In April 2020, 37 states implemented orders, except North Dakota, Nebraska, and
Iowa. I define a fund to be locked down once its zip code had a stay-at-home order issued.
The second measure of change in face-to-face communication is the foot traffic activity at the
zip code level obtained from the SafeGraph Places Patterns database. The dataset provides an
hourly number of visits to 3.6 million commercial points-of-interests (POIs) from over 45 million
mobile devices in the US. The sample consists of a panel of opt-in, anonymous smartphones,
which is well balanced across US demographics and geographies, covering roughly 10% of the US
population. The data has been used in recent Finance research including Liu and Lu (2021), Ng,
Yu, and Yu (2021), and Bai and Massa (2022).
I compute the monthly footprint activities at the zip code level by taking the sum of the
number of visits to all POIs within a zip code at the monthly frequency. For some zip codes with
missing footprint data, I use the footprint activities of the nearest zip code. Figure 2.4 Panel (a)
plots the average of the total footprint traffic aggregated across all zip codes of fund managers.
The sudden drop in March, with the most significant dip in April 2020, is noticeable. To measure
the change in foot traffic, I calculate the percentage change of total footprint activities between
the 2019 average and the monthly footprint activities at the zip code level. Table 2.2 reports the
28

summary statistics from March to June 2020. The median footprint percentage changes are −25%,
−63%, −50%, and −41% for each month, respectively.
Figure 2.4 Panel (b) shows the distribution of lockdowns across funds and time, when the
footprint activities are used to proxy for local in-person activities. I set −30% as the threshold to
define a lockdown given that the median footprint percentage changes in March 2020 is −25%.
A little less than half of the sample funds experienced such a drop in March 2020, and almost all
sample funds had a drop by April 2020. For the main analyses, I define a fund manager’s zipcode
location to be locked down once its footprint activities drop by more than 30%. For the funds
with several zip codes, I consider them to be locked down if at least one of the zip codes is locked
down. For the analyses at the stock level, I define a firm’s headquarters area as being locked down
once its local area within 100 miles experiences a drop in footprint activity of more than 30%.
The final variable that I use to measure the decrease in face-to-face communication is the number of Covid cases, which picks up individuals’ endogenous decision to stay at home. Monthly
state-level case counts scaled by the number of local populations are obtained from the Centers
for Disease Control and Prevention (CDC) COVID-19 Data Tracker. I use the case counts of a
lead manager’s state for the funds with multiple locations.
As expected, the three measures are highly correlated. The correlation coefficient is 0.89
for stay-at-home orders and the percentage change of footprint activities, 0.71 for stay-at-home
orders and Covid case counts, and 0.77 for the percentage change of footprint activities and Covid
case counts. This suggests that the three measures account for the same shock caused by the
outbreak of the pandemic.
29

2.4 Impact of Lockdowns on Investment Decisions
This section investigates the adverse impact of lockdowns on fund managers’ investment decision
making on local stocks. After providing suggestive evidence based on the tests with portfoliolevel returns, I investigate the change in fund managers’ investment timings at the stock level to
identify the causal role of face-to-face meetings in generating local informational advantages.
2.4.1 Portfolio-level return
This section investigates the change in fund managers’ portfolio-level investment performance
on local stocks relative to non-local stocks after lockdowns. I divide each portfolio into a local and
distant portion based on a 100-mile local threshold. Monthly returns are calculated separately for
local and distant portions of every fund as follows:
R
L(D)
i,t =
Li,t
X
(Di,t)
j=1
w
L(D)
ij,t × rj,t (2.1)
where RL
i,t (RD
i,t) is the monthly return of fund i in month t on local (distant) holdings, Li,t
(Di,t) is the number of local (distant) stocks held by fund i in month t, w
L
ij,t (w
D
ij,t) are the rescaled
(to sum to one) portfolio value weights applied to the fund i’s local (distant) holdings, and rj,t is
the return of stock j in month t.
For rj,t, I employ a fund’s benchmark excess returns based on Morningstar style, and the
characteristic-adjusted returns as in Daniel, Grinblatt, Titman, and Wermers (DGTW) (1997),
which I henceforth call DGTW returns. To obtain DGTW returns, I sort all stocks in CRSP into
size quintiles, then sort stocks into book-to-market quintiles within each size quintile, and finally
30

sort the stocks into momentum quintiles within each group. Benchmark portfolios are formed by
value-weighting the stocks within each of these 125 groups. Stocks are matched with one of the
125 portfolios based on the three characteristics of the previous month. DGTW return is obtained
by subtracting the return of the matched portfolio from the individual stock return.
I investigate how lockdowns affected fund managers’ investment performance on local stocks
by running the following regression:
R
L,D
i,t = β0 + β1Locali + β2Lockdowni,t + β3Locali × Lockdowni,t
+ F undF Ei + T imeF Et + ϵi,t
(2.2)
where RL
i,t(RD
i,t) is the value-weighted monthly return of the local (non-local) portion of fund
i in month t. Locali
is a dummy variable equal to one for the local portfolio. Lockdowni,t is a
dummy variable equal to 1 if the zip code of fund i had a stay-at-home order implemented in
month t, or if its footprint activity dropped by more than 30% relative to the 2019 average. Fund
fixed effects control for time-invariant features of fund managers, and year-month fixed effects
control for the macro level shock. Standard errors are clustered by fund. Figure 2.5 confirms the
parallel trend assumption.
Table 2.3 reports the regression results. The coefficient of interest is that of the interaction
term between local and lockdown dummies. I find that funds on average had worse performance
on local stocks relative to non-local stocks during lockdowns. Specifically, a fund’s local benchmark excess return (DGTW return) decreased by 0.61 (0.42) percentage points relative to their
non-local returns when local in-person interactions are proxied by stay-at-home orders, and by
0.7 (0.5) percentage points when they are proxied by footprint activities on average. The magnitudes are non-trivial as the difference between local and distant portfolios became about three
31

times bigger than in the pre-lockdown period. These results hold when the lockdown dummy
is replaced by the continuous percentage change of footprint activities and the number of Covid
cases.
Overall, the results provide suggestive evidence that the loss of face-to-face interactions during lockdowns disrupted the flow of information from local firms to fund managers.
2.4.2 Investment timings
As illustrated in Section 2.2, the portfolio-level tests in the previous section provide only suggestive evidence as the effect of the changes in underlying stock returns with lockdowns may
confound the results. Therefore, this section runs stock-level tests to isolate the effect of investment timings holding fixed all other aspects of the investment, before and after lockdowns.
Specifically, I calculate each fund’s monthly investment returns for every position taken after
January 2019 using the dollar value of holdings, assuming that positions are bought or sold at the
previous month’s end price. This investment return reflects the timing of the trades. Specifically,
the investment return is proportional to how much a fund has changed its position in conjunction with the stock return. If a fund does not trade and holds the same position as in the previous
month, the investment return would be the same as the stock return. For traded positions, the
return becomes higher than the stock return for increases in position before the stock price increase and lower for decreases in position before the stock price increase. For the initiation of a
new position without a position in the previous month, adjustments are made to set the change in
holding percentage to 100%. After calculating each fund’s monthly returns on each stock, I compute three-month investment returns by cumulating returns in a rolling window at the monthly
32

frequency. I include fund-stock pairs if a fund holds any position in stock for at least one month
during the sample period. For months with no holdings for a pair, either because a fund made
a new investment decision after January 2019 or because a fund has liquidated the position, the
investment return is set to 0%.
Table 2.4, Panel (a) presents summary statistics of investment returns, where the sample period is from January 2019 to September 2020. To isolate time-series changes in fund managers’
investment timing decisions from all other aspects of the investment, I run the following regression:
Returni,j,t = β0 + β1Locali,j + β2P ostLockdownj,t + β3Locali,j × P ostLockdownj,t
+ F undF Ei + StockF Ej × P ostLockdownF Ej,t + ϵi,j,t
(2.3)
where Returni,j,t is three-month investment return of fund i on stock j in month t, Locali,j
is a dummy equal to 1 if the distance between fund i and firm j is shorter than 100 miles,
P ostLockdownj,t is a dummy indicating a post lockdown period based on the footprint activities in the local area of firm j’s headquarters. Fund fixed effects control for time-invariant traits
of fund managers, and Stock×Post-lockdown fixed effects absorb the effect of each stock’s average return over the short time span before and after the firm area is locked down. Standard errors
are clustered by stock, fund, and time. Figure 2.6 checks for the parallel trend assumption.
Table 2.4 Panel (b) shows the regression results. The coefficient of interest is that of the
interaction term, which is statistically significantly negative in all specifications. Specifically,
33

Column (2) with Stock×Post-lockdown and Fund fixed effects show that compared to the prelockdown period, three-month investment returns on a specific stock by local fund managers are
lower by 0.04 percentage points during lockdowns compared to distant managers investing in
the same stock. This translates to an economic loss of $0.4 million per portfolio on average on an
annual basis. Column (4) shows that the same pattern continues until the end of 2020, although
the magnitude drops by 30%, which is consistent with the idea that investors still refrained from
engaging in face-to-face social interactions during the pandemic even after the official lockdown
orders were lifted.
The results suggest that fund managers’ poor local performance during lockdowns is driven
by their deteriorated investment timings and not by the deteriorated stock returns of the firms
during lockdowns.
2.5 Channel Analysis
This section explores the channels through which face-to-face interactions matter for informational advantages. I examine fund managers’ buy and sell decisions and the heterogeneity across
stocks with different informational environments. I then consider the possibility of the results
being driven by a decrease in information from alternative sources, the information spread within
fund families and fund managers’ public information-seeking activities.
34

2.5.1 Buy versus sell decisions
This section explores the investment timings of buy- and sell-orders separately to uncover the
nature of the information transmitted in face-to-face settings. Specifically, I examine the next
period’s characteristic-adjusted returns of traded stocks in the following regression:
StockDGTWj,t+1 = β0 + β1Locali,j + β2P ostLockdowni,t
+ β3Locali,j × P ostLockdowni,t + StockDGTWj,t
+ StockF Ej × F undF Ei + T imeF Et + ϵj,t
(2.4)
where the dependent variable is the next month’s DGTW return of stock j which is either
bought or sold by fund i in month t. The returns of sold stocks are multiplied by −1 so that the
negative next-period stock returns indicate poor investment timings. StockDGTWj,t controls
for the stock’s return in the month when the stock was traded. Stock×Fund fixed effects control
for any time-invariant stock-fund pair specific unobservables, which allows for the comparison
of investment timings on stock j by fund manager i before and after a fund manager is locked
down. Standard errors are clustered by fund and stock. Figure 2.7 checks for the parallel trend
assumption.
I begin with running the regression while only including the positions in which fund managers increased holdings during the sample period. Table 2.5 Column (1) presents the results.
The negative coefficient on the interaction term indicates that, after controlling for the stock
return of the current period, the next period return of a local stock that is added to a portfolio
becomes lower by 0.006 percentage points, on average, relative to distant stocks after lockdowns,
35

which is equivalent to 2.1% of the pre-lockdown DGTW returns of local stocks. This result suggests that fund managers’ investment timings on buy-orders of the same stock deteriorated after
lockdowns.
Next, I compare fund managers’ investment timings on buy- and sell-orders by including
both the positions with increased and decreased holdings, and interacting the DiD interaction
term with a dummy Buy which is 1 for bought stocks and 0 for sold stocks. Table 2.5 Column
(2) presents the result. The negative coefficient on the triple interaction term indicates that buy
decisions were more negatively impacted by lockdowns relative to sell decisions. Specifically,
compared to sold stocks, bought stocks’ drop in the (signed) next-month return was bigger by
0.012 percentage points on average for local positions. This suggests that the information transmitted in face-to-face settings is more likely to be of positive content, and fund managers use the
information advantageously when adding local positions to their portfolios.
I explore whether changes in fund managers’ aggressiveness or activeness in making trading
decisions drive these results. To examine if fund managers’ aggressiveness of trades on the stock
changed after lockdowns, I use the log of the dollar value of trading amounts as the dependent
variable in the regression specification of Equation (2.4), controlling for the previous month’s
position in dollar amounts. Table 2.5 Column (3) presents the result with buy-orders and Column
(4) presents the result that compares buy- to sell-orders. Insignificant results in both columns
suggest that during lockdowns, fund managers were equally aggressive in making buy and sell
decisions as in the pre-lockdown period.
Next, I investigate whether the managers traded as much during lockdowns as they did in the
pre-lockdown period. I do so by using monthly portfolio-level turnover ratios as the dependent
variable. Each fund’s overall, local, and non-local monthly turnover ratios are calculated as the
36

minimum of aggregate purchase and sale divided by the monthly fund assets following Yan and
Zhang (2009). As the regression is at the portfolio level, fund and year-month fixed effects are
now included. Table 2.5 Column (5) presents the results on the overall turnover ratio, and Column (6) presents those on local and non-local turnover ratios. Again, statistically insignificant
results suggest that managers did not significantly change their activeness in making investment
decisions during lockdowns.
Together, the results suggest that fund managers’ investment timings on local stocks, especially when buying them, deteriorated after lockdowns. The results are not driven by the change
in their aggressiveness or activeness in making trading decisions. These suggest that fund managers utilize the positive information gathered through face-to-face social interactions to execute
profitable trades when adding local stocks into their portfolios.
2.5.2 Heterogeneity across stock informational environment
If managers lost access to the information profitable for buy-orders during lockdowns, the results
would be pronounced for stocks that are informationally sensitive. I explore the cross-sectional
heterogeneity of the results in the previous section across stocks with different characteristics.
Specifically, I examine whether the adverse impact of lockdowns on buy decisions arises bigger
for stocks with less transparent informational environments.
I focus on several stock characteristics that indicate how informationally sensitive a stock
is. Sample stocks are divided into two groups based on the 2019 median of 1) dollar value of
trade size 2) trading volume 3) idiosyncratic volatility, the standard deviation of the residuals
when daily stock returns are regressed on Fama-French three factors 4) Amihud illiquidity 5)
37

institutional ownership obtained from Thomson Reuters 13F institutional ownership 6) analyst
forecast dispersion, the standard deviation of analyst earnings forecast obtained from I/B/E/S 7)
analyst coverage from I/B/E/S 8) media coverage obtained from Ravenpack 9) total asset, and 10)
S&P500 index inclusion.
After categorizing stocks based on the characteristics, I run the regression in Table 2.5 Column (2) that compares buy and sell decisions, but additionally interacting the triple interaction
term with a dummy variable indicating if a stock has above median characteristic. This is an
extension of the regression specification in Equation (2.4) to include a four-way interaction term,
Local×Post-Lockdown×Buy×AboveMedian. The coefficient on the four-way interaction term
indicates how the impact of lockdowns on buy decisions relative to sell decisions, on local stocks
relative to distant stocks, varies across firms with different informational environments. Standard
errors are clustered by fund and stock.
Figure 2.8 presents the point estimates and the 90% confidence intervals of the interaction
terms in ten different regressions on each stock characteristic. The point estimates are significantly negative or insignificantly positive at the 90% confidence level, partially supporting the
idea that fund managers’ investment timings on informationally sensitive stocks were impaired
to a greater extent. Specifically, the statistically significant result suggests that the adverse impact of lockdowns on buy decisions relative to sell decisions arises bigger for the stocks with
large pre-lockdown trade sizes. As a large trade size indicates that fund managers were making
informed trades, the result indicates that fund managers lost access to superior information due
to curtailed face-to-face communication during lockdowns.
Moreover, the adverse effect of lockdowns on buy decisions arises bigger for the stocks with
high idiosyncratic risks, the stocks that possess firm-specific uncertainty that fund managers
38

cannot insure against. Also, the result arises bigger for small firms with less publicly available
information. In the same context, the effect arises bigger for the stocks with high analyst forecast
dispersions, the stocks with an unobservable underlying value that induces a large dispersion
in the belief in the prospect of a firm. Finally, the result arises bigger for the stocks with high
Amihud illiquidity, which implies higher transaction costs and illiquidity risks.
Overall, these suggest that fund managers utilize the positive information gathered through
face-to-face social interactions to execute profitable trades on the stock with a less transparent
informational environment.
2.5.3 Information flow within fund families
Now I consider the possibility that previous results are driven by a decrease in information from
another source that may also have been affected by lockdowns: information flow within fund
families documented in Cici, Jaspersen, and Kempf (2017).∥ As fund managers worked from home
after lockdowns, during which communications among colleagues may not have been as smooth
as in the pre-lockdown period, I investigate if the disrupted information flow within organizations
is driving the results.
I first examine if lockdowns indeed disrupted the flow of information within mutual fund
families. To do so, I employ the Speed of Information Diffusion (SID) measure of Cici, Jaspersen,
and Kempf (2017), which traces the sequence of trades within fund families after a stock is first
introduced by one of the affiliated fund managers. Specifically, the speed of information diffusion
of each stock initiation in a family is defined as follows:
∥
I thank an anonymous referee for suggesting exploring this channel.
39

IDf,s,q =
If,s,q − 1
If,s,q + Jf,s,q − 1
(2.5)
where If,s,q is the number of funds in family f that initiates a position in stock s that is
not already held by any fund in the family in quarter q, Jf,s,q is the number of funds in the
family that follow later during an information interval. The information interval starts when
the initial stock purchase happens and ends when the initiating manager liquidates the stock.
Information diffusion is observed only when at least two funds from the family trade stock s
(I+J >1), and IDf,s,q is bounded between zero and one. A larger value indicates a faster speed of
information diffusion. The speed of information diffusion at the family level, SIDf,q, is computed
by averaging IDf,s,q corresponding to information intervals, the last purchase of which happens
during the last four quarters including quarter q.
To capture different speeds of information diffusion among managers with the same versus
different investment styles, I further compute SIDW ithin and SIDAcross following Cici, Jaspersen,
and Kempf (2017). SIDW ithin measures SID among affiliated managers with the same investment style, which is computed by averaging ID across all styles within a family. SIDAcross
measures SID across different investment styles, which is computed using the portfolio holdings aggregated for each style and the sequence of trades across the aggregated portfolios of all
styles.
Table 2.6 Panel (a) shows the t-test results that compare SID before and after lockdowns. Because SID is measured every quarter, post-lockdown indicates the quarters starting from 2020
Q2. The results suggest that the speed of information flow within a family statistically significantly decreased by 3.65% on average after lockdowns. Panel (b) presents the results with fund
40

family fixed effects, which control for time-invariant fund family characteristics. Again, the significantly negative coefficients indicate that the flow of information within fund families, both
within and across styles, was disrupted during lockdowns.
Given the interrupted within-organizational information flow during lockdowns, I test if this
is driving the results on fund managers’ impaired investment timings during lockdowns in Section
2.4.2. If this were the case, the results could arise either way: On the one hand, the adverse impact
of lockdowns would be pronounced for the managers from relatively high SID fund families if
they suffer from the decreased SID during lockdowns. On the other hand, the organizational
structure of high SID may enable them to better cope with such disruption. To test this, I run
the stock-level regression on investment timings as in Equation (2.3), additionally interacting the
interaction term with a dummy HighSID that indicates the funds from fund families with above
median SID during 2019.
Table 2.6 Panel (c) presents the results. The coefficient of interest is that of the three-way
interaction term, Local×Post-lockdown×High SID. Overall, the statistically insignificant results
suggest that the disrupted internal information is not driving the results on the deteriorated investment timings.
Next, I consider another related possibility that fund managers obtain information on distant stocks that are local to their colleagues from the same family. If this were the case, and if
lockdowns interrupted communication among colleagues, fund managers’ investment timings
on the stocks that are local to their distant colleagues would deteriorate after lockdowns. To test
the idea, I run the same regression on investment timings but replacing the dummy Local with
a dummy Branch, which is 1 for the stocks headquartered in the states where a fund family
branch exists, and 0 for the stocks headquartered in the states without a branch. The regression
41

compares the investment timing on the same stock by two fund managers, those from a family
with and without a branch near the firm headquarters.
Table 2.6 Panel (d) presents the result. Again, the statistically insignificant results in all specifications suggest that the information from distant colleagues within families is not critical in
generating informational advantages.
Combined, the results suggest that the change in within-organizational information flow is
not critical enough to drive the results on the adverse impact of lockdowns. Instead, they are
in line with the idea that fund managers’ investment decisions deteriorated due to the disrupted
information flow from local firms during lockdowns.
2.5.4 Public information seeking
This section explores another important source of information for fund managers, public information.∗∗ As Dyer (2021) documents using EDGAR log files, institutional investors acquire more
public information on local stocks to make superior trading decisions. I explore if fund managers’
public information seeking on local stocks changed during lockdowns to drive the results on the
adverse impact of lockdowns.
Because EDGAR log files are unavailable for early 2020 during which Covid lockdowns were
implemented, I alternatively rely on I/B/E/S stock analyst recommendations as a proxy for public
information. The data provides investment recommendations for all stocks tracked by sell-side
analysts in the range of 1 for "strong buy" to 5 for "strong sell". I follow Kacperczyk and Seru (2007)
to examine how much of the average percentage changes in a fund’s holdings can be attributed to
∗∗I thank an anonymous referee for suggesting exploring this channel.
42

changes in analysts’ recommendations. Specifically, I run the following cross-sectional regression
for each fund f and quarter q using all stocks s=1 to n in the fund’s portfolio:
%∆Holdf,s,q =β0,q + β1,q∆Res,q−1 + β2,q∆Res,q−2 + β3,q∆Res,q−3
+ β4,q∆Res,q−4 + ϵf,q, ∀s = 1 · · · n
(2.6)
where %∆Holdf,s,q denotes a percentage change in the number of holdings or dollar value
of holdings of stock s held by fund f from quarter q − 1 to q, ∆Res,q−p measures a change in the
recommendation of the consensus forecast of stock s from quarter q − p − 1 to quarter q − p,
and p = 1, 2, 3, 4 is the number of lags of the forecast. %∆Holdf,s,q is set to 100% when a new
stock position is initiated. Reliance on Public Information (RP I) is the unadjusted R2 of the
regression. I denote the measure as RP I if the number of holdings is used and as RP Idollar if the
dollar value of holdings is used to compute %∆Holdf,s,q.
Table 2.7 Panel (a) presents the t-test results that compare RP I(RP Idollar) before and after
lockdowns, where post-lockdown refers to the quarters starting from 2020 Q2. RP I (RP Idollar)
increased by 2.7% (5.3%) on average with a t-statistic of 1.38 (2.21), suggesting that fund managers
increased their overall reliance on public information during lockdowns.
To examine if fund managers increased their reliance on public information to a greater extent
on local stocks than on distant stocks, I run DiD regressions at the fund level using RP I calculated
separately for local and distant holdings as the dependent variable. Table 2.7 Panel (b) presents
the result. Although the result is statistically insignificant when the number of holdings is used,
the significantly positive coefficient in Column (2) with the dollar value of holdings suggests that
managers relied more on analysts’ recommendations for local investments after lockdowns.
43

In sum, the results suggest that fund managers’ public information seeking did not decrease
during lockdowns, particularly for local stocks for which they lost the face-to-face channel. Together with previous results, this suggests that managers were not able to mitigate the adverse
impact of lockdowns by seeking more public information.
2.6 Impact of Lockdowns on Local Bias
As the results so far suggest that Covid lockdowns adversely impacted mutual fund managers’
investment decisions on local stocks, I next explore how fund managers’ local biases changed
during lockdowns. If face-to-face interaction is an important factor in determining fund managers’ preferences for local stocks, the absence of such interactions during lockdowns will cause
fund managers’ local biases to converge.
To gauge the degree to which a fund manager invests locally, I construct a local bias measure following Coval and Moskowitz (2001). First, I compute the fraction of the portfolio’s assets
invested in stocks located within 100 miles. As funds differ in the density of available investments in their local area, I compute the fraction of the market of available investments within
100 miles.†† Local bias is defined as the difference between the two fractions, which represents
the degree to which a manager invests locally in excess of the market portfolio.
The summary statistics of the fraction of the assets invested in local stocks and the local bias
for the pre- and post-pandemic periods are presented in Table 2.1. Funds on average allocate 12%
of their assets to local stocks. The distribution is skewed to the right with some funds showing
a particularly high preference for local stocks. The average local bias during the entire sample
††Stocks held by at least one fund in the sample are considered to be the universe of assets available for investment.
44

period is 2.85%, and the median local bias is 1.55%. The median value is comparable to the value
documented in Bernile, Kumar, Sulaeman, and Wang (2019), which are 2.63% during 1996-1999
and 1.41% during 2000-2008.
To account for the different preferences toward local stocks before the pandemic, I categorize
sample funds into three groups based on their median local bias in 2019. The funds in the lowest
and the highest terciles are similarly located across the US. For the funds that have the least prepandemic local bias (T1), 36% are in Massachusetts, 20% are in New York, 9% are in California,
and 5% are in Pennsylvania and Texas. For the funds with the greatest pre-pandemic local bias
(T3), 24% are in New York, 15% are in California, 8% are in Illinois, 6% are in Pennsylvania, and
3% are in Texas.
Figure 2.9 plots the average local bias of each tercile. The large distances between the green
long-dashed line for the most biased funds and the other two lines indicate a skewed distribution.
The slight downward trend of the green long dashed line (T3) and the slight upward trend of the
blue line (T1) after March 2020 provide preliminary evidence that the difference in the local bias
among fund managers decreased during lockdowns, suggesting that face-to-face interaction is
one of the important factors affecting their local bias. I further examine the change in fund
managers’ local bias in the following regression specification:
LocalBiasi,t = β0 + β1Lockdowni,t + β2Lockdowni,t × T2i + β3Lockdowni,t × T3i
+ F undF Ei + T imeF Et + ϵi,t
(2.7)
45

where the dependent variable is the local bias of fund i in month t. Lockdowni,t is a dummy
variable equal to 1 if the zip code of fund i has a stay-at-home order implemented in month t,
or if its footprint activity dropped by more than 30% relative to the 2019 average. The lockdown
dummy is interacted with a categorical variable that divides funds into terciles based on their
local bias in 2019. Standard errors are clustered by fund. Figure 2.10 confirms the parallel trend
assumption.
Table 2.8 reports the results. As the funds that have the least pre-pandemic local bias (T1) are
the baseline category, the statistically significantly positive coefficient on the lockdown dummy
indicates that the least biased funds increased their local holdings during lockdowns. Specifically,
they increased local bias by 17-19% relative to the pre-lockdown median local bias.
On the other hand, the statistically significantly negative coefficient on the interaction term
between the lockdown dummy and T3 indicates that during lockdowns, the average change in
local bias of the most biased funds in T3 is lower than that of the least biased funds in T1 by about
0.8 percentage points.
Combined, the results suggest that curtailed face-to-face communication reduced the difference in the preference for local stocks across fund managers. This indicates that face-to-face
interaction is an important factor in explaining fund managers’ preference for local stocks.
2.7 Social Index
This section explores whether the fund managers who enjoyed a greater local advantage before
lockdowns were impacted by lockdowns to a greater extent. To do so, I employ a measure of social
index to investigate whether the previous results on the investment performance in Section 2.4
46

and the results on local bias in Section 2.6 are pronounced for the fund managers in the regions
with strong social ties.
Following Hasan, HOI, Wu, and Zhang (2017) and Kang, Stice-Lawrence, and Wong (2021), I
exploit a social index measure from the Northeast Regional Center for Rural Development at the
Pennsylvania State University to proxy for the strength of investors’ social ties within their local
communities. I use the variable ASSN 2014, which is the number of ten types of social organizations for all US counties in 2014, which include nonprofit organizations; social organizations
such as sports clubs, public golf courses, bowling, and fitness centers; and associations with a
professional, business, political, religious, or other orientation.
After dividing funds into three groups based on the social index, I run the regression on the
local portfolio returns as in Equation (2.2) and on the local bias as in Equation (2.7) separately for
the lowest and the highest terciles. Table 2.9 Panel (a) shows the result on portfolio-level returns.
The coefficients of the interaction terms are bigger in magnitude and statistically significant for
the high-index group while those for the low-index group are statistically insignificant. Similarly,
the results on local bias in Panel (b) shows a larger magnitude of the coefficients on the interaction
terms for the high-index group. The results show that the adverse impact of lockdowns was
pronounced for the funds in the regions with strong social ties before lockdowns.
Together, the results suggest that the fund managers who enjoyed a greater local advantage
before lockdowns were more significantly impacted by lockdowns. This supports the idea that
fund managers rely on face-to-face social interactions to create an informational edge on local
stocks.
47

2.8 Conclusion
In this study, I investigate whether face-to-face social interaction is important for mutual fund
managers in obtaining value-relevant information on local stocks. By setting 100 miles as the local
threshold, employing the exact residential location of fund managers, and exploiting COVID-19
lockdowns in early 2020, I run DiD regressions to explore how the curtailed in-person activities
adversely affected fund managers’ trading behaviors on local stocks relative to non-local stocks.
I find that during lockdowns, mutual fund managers’ performance on local stocks declined
relative to non-local stocks because the timing of their trades deteriorated, particularly for buyorders of informationally sensitive stocks. The results are not driven by the change in the underlying stock returns nor by the change in the aggressiveness or activeness of trades. The results
are neither driven by a decrease of information from alternative information sources, information flow within fund families and fund managers’ reliance on public information. Additionally, I
document that fund managers’ preferences for local stocks converged during lockdowns. Finally,
I show that the results are more pronounced for the fund managers who enjoyed a greater local
advantage before lockdowns, those in the regions with strong social ties.
Combined, the results highlight that, even with advanced communication technologies, the
sharing of comprehensive stock information cannot be sustained without continuous face-to-face
social interactions.
48

Figure 2.1: Example of different investment performance on the same stock
This figure illustrates how different investment decisions can lead to varying returns for the
same stock. The blue solid line represents the three-month return of a stock. The green longdashed and red short-dashed lines depict the three-month returns of two funds that invested in
the stock. These returns are calculated as the cumulative monthly percentage changes in the
dollar value of holdings, assuming that new positions are bought or sold at the previous month’s
end price. In this example, both funds initially held the same position in the stock but executed
additional trades at different times, resulting in different time series of returns. Specifically, Fund
1 doubled its position during January 2020, after which the stock had a positive return at the end
of the month. Consequently, Fund 1 achieved a higher return than the stock itself due to welltimed trades, while Fund 2, which did not adjust its position, had the same return as the stock.
However, when Fund 1 sold its position during April 2020 after which the stock’s return was
positive, Fund 1’s return becomes lower than that of the stock due to a timed trade. In contrast,
when Fund 2 doubled its position during May 2020, the stock experienced significantly positive
returns. Consequently, Fund 2 achieved a higher return than both the stock and Fund 1.
49

Figure 2.2: Mutual fund manager location
The figure depicts the geographical distribution of mutual fund managers across the US at the
state and county levels. The circles in Panel (a) represent the number of fund managers in each
state. The sizes of New York and Massachusetts are scaled down by half. Panel (b) shows the
location at the county level, red indicating a high percentage of managers.
(a) State level
(b) County level
50

Figure 2.3: Stay-at-home orders
The figures show the adoption of stay-at-home orders across the US in March and April
of 2020. Red indicates the state with orders, blue with no orders, and grey indicates the states with no fund managers in the sample. In March 2020, lockdown orders were implemented in 27 states out of 40 states in which fund managers are located.
In April 2020, 37 states had the orders except for North Dakota, Nebraska, and Iowa.
(a) March 2020
(b) April 2020
51

Figure 2.4: Footprint activities
The figure shows the footprint activities in the location of sample fund managers. Panel (a) plots
the average monthly total footprint activities across all zip codes of mutual fund managers. Panel
(b) shows the distribution of the footprint change across funds and time by setting −30% as the
threshold to define lockdowns. The x-axis indicates time, and the y-axis indicates funds. Red
means the monthly footprint activities of a fund’s zip code dropped by more than 30% relative to
the 2019 average, and blue indicates a change smaller than that.
(a) Aggregate footprint activities
(b) Change in footprint activities across funds and time
52

Figure 2.5: Impact of lockdowns on local portfolio return
This figure depicts the parallel trend for the regression results in Table 2.3. The figure plots
the point estimates of γs and 95 percent confidence intervals calculated using standard errors
clustered by fund in the following regression:
R
L,D
i,t = β0 + β1Locali +
X
t+3
s=t−5
(βsEventi,s + γsEventi,s × Locali)
+ F undF Ei + T imeF Et + ϵi,t
where R
L,D
i,t is the benchmark excess return, Locali
is a dummy equal to one for the local portion
of a portfolio, and Eventi,s is a time indicator relative to the lockdown month in which footprint
activities drop by more than 30% relative to the 2019 average. The coefficients are compared to
that of the month prior to the lockdown.
53

Figure 2.6: Impact of lockdowns on local stock-level return
This figure depicts the parallel trend for the stock-level regression results in Table 2.4. The figure
plots the point estimates of γs and 95 percent confidence intervals calculated using standard errors
clustered by stock, fund, and time in the following regression:
Returni,j,t = β0 + β1Locali,j +
X
t+3
s=t−5
(βsEventj,s + γsEventj,s × Locali,j )
+ F undF Ei + StockF Ej × P ostLockdownF Ej,t + ϵi,j,t
where Returni,j,t is the three-month investment return of fund i on stock j in month t, Locali,j
is a dummy equal to 1 if the distance between fund i and firm j is shorter than 100 miles, and
Eventj,s is a time indicator relative to the lockdown month in which footprint activities in the
local areas of firm j’s headquarters drop by more than 30% relative to the 2019 average. The
coefficients are compared to that of the month prior to the lockdown.
54

Figure 2.7: Impact of lockdowns on the investment timings of buy orders
This figure depicts the parallel trend for the regression results in Table 2.5 Column (1). The figure
plots the point estimates of γs with 95% confidence intervals calculated using standard errors
clustered by fund and stock in the following regression:
StockDGTWj,t+1 = β0 + β1Locali,j +
X
t+3
s=t−5
(βsEventi,s + γsEventi,s × Locali,j )
StockDGTWj,t + StockF Ej × F undF Ei + T imeF Et + ϵj,t
where StockDGTWj,t+1 and StockDGTWj,t is the next and current month’s DGTW return
of stock j that is bought by fund i at month t. Other variables are defined in Figure 2.5. The
coefficients are compared to that of the month prior to the lockdown.
55

Figure 2.8: Heterogeneity across stock informational environment
This figure reports the results of the regression in Table 2.5 Column (2) that compares buy and
sell decisions, but additionally interacting the triple interaction term with a dummy variable indicating if a stock has above median characteristic. The figure plots the point estimates and the 90%
confidence intervals of four-way interaction terms, Local×Post-Lockdown×Buy×AboveMedian.
Sample stocks are divided into two groups based on the 2019 median of 1) dollar value of trade
size 2) trading volume 3) idiosyncratic volatility, the standard deviation of the residuals when
daily stock returns are regressed on Fama-French three factors 4) Amihud illiquidity 5) institutional ownership obtained from Thomson Reuters 13-F Filings 6) analyst forecast dispersion, the
standard deviation of analyst earnings forecast obtained from I/B/E/S 7) analyst coverage from
I/B/E/S 8) media coverage obtained from Ravenpack 9) total asset, and 10) S&P500 index inclusion.
Standard errors are clustered by fund and stock.
56

Figure 2.9: Local bias
The figure plots the average local bias of sample funds divided into three groups based on their
pre-pandemic local bias. Local bias is defined as the difference between the asset fraction of
holdings within 100 miles and the fraction of the market of available investments within 100
miles.
57

Figure 2.10: Impact of lockdowns on local bias
The figure depicts the parallel trend for the regression in Table 2.8. The figure plots the point
estimates of the interaction term (γ2,s) with 95% confidence intervals calculated using standard
errors clustered by fund in the following regression:
LocalBiasi,t = β0 +
X
t+3
s=t−5
(βsEventi,s + γ1,sEventi,s × T2i + γ2,sEventi,s × T3i)
+ F undF Ei + T imeF Et + ϵi,t
where LocalBiasi,t is the local bias of fund i in month t, Eventi,s is a time indicator relative to the
lockdown month in which footprint activities drop by more than 30% relative to the 2019 average,
which is then interacted with a categorical variable that divides funds into terciles based on their
local bias in 2019. The coefficients are compared to that of the month prior to the lockdown.
58

Table 2.1: Mutual fund characteristics
This table reports the characteristics of actively-managed US equity mutual funds in the sample
for the pre-pandemic period (January 2019 - February 2020) in Panel (a) and the post-pandemic
period (March 2020 - June 2020) in Panel (b).
Variable N Mean Median St. Dev. Min Pctl(25) Pctl(75) Max
Panel (a) January 2019 - February 2020
Number of managers 12,942 2.76 2 2.47 1 1 3 24
Number of regions 12,942 1.20 1 0.65 1 1 1 7
Holding distance (mile), p10 12,942 182.58 150.12 177.81 0.60 49.93 237.07 1,248.44
Fund AUM ($bil) 12,942 1.49 0.40 2.44 0.001 0.09 1.54 9.51
Fund assets in local stocks (%) 12,942 11.98 9.13 9.89 0.81 4.52 16.83 59.68
Local (100 miles) bias (%) 12,942 2.84 1.54 7.66 −21.60 −1.13 5.96 27.48
Number of holdings 12,942 84.96 63 67.80 20 42 98 331
Number of local holdings 12,942 10.94 6 15.142 1 3 12 155
Panel (b) March 2020 - June 2020
Number of managers 3,976 2.77 2 2.48 1 1 3 24
Number of regions 3,976 1.20 1 0.64 1 1 1 7
Holding distance (mile), p10 3,976 188.13 150.76 184.76 0.62 49.18 247.41 1,475.31
Fund AUM ($bil) 3,976 1.38 0.32 2.38 0.002 0.08 1.34 9.51
Fund assets in local stocks (%) 3,976 11.94 8.94 10.20 0.81 4.37 16.49 60.50
Local (100 miles) bias (%) 3,976 2.95 1.63 7.61 −22.69 −1.07 6.21 27.48
Number of holdings 3,976 85.25 62 69.19 20 42 98 331
Number of local holdings 3,976 11.08 6 15.53 1 3 13 148
Table 2.2: Change in footprint activities (%) relative to the 2019 average
The table reports the summary statistics of the percentage change in the footprint activities from
March to June 2020, which is the percentage change between the average footprints of 2019 and
the monthly footprints at the zip code level.
Time Mean Min p10 p25 p50 p75 p90 Max
March 2020 −24.38 −77.08 −41.33 −32.79 −24.59 −17.02 −9.20 339.93
April 2020 −62.01 −96.77 −78.87 −71.62 −63.07 −53.17 −43.20 −11.92
May 2020 −49.39 −95.52 −72.13 −62.37 −50.37 −37.78 −24.52 38.69
June 2020 −39.85 −97.85 −66.99 −53.79 −40.52 −26.45 −12.44 136.94
59

Table 2.3: Impact of lockdowns on local portfolio return
This table presents the regression results about the impact of lockdowns on portfolio-level return. Lockdown is defined in four ways:
1) a dummy variable equal to 1 if the zip code of a fund had a stay-at-home order implemented 2) a dummy variable equal to 1 if the
footprint activity dropped by more than 30% relative to the 2019 average 3) percentage change in footprint activities relative to the
2019 average 4) state-level Covid case counts scaled by the number of the local populations. Standard errors are clustered by fund.
Dependent variable:
Benchmark excess return (%) DGTW return (%)
(1) (2) (3) (4) (5) (6) (7) (8)
Local dummy −0.236∗∗∗−0.227∗∗∗−0.226∗∗∗ −0.268∗∗∗ −0.116∗∗∗−0.105∗∗∗−0.104∗∗∗ −0.127∗∗∗
(0.039) (0.039) (0.039) (0.039) (0.034) (0.034) (0.034) (0.033)
Lockdown order dummy 1.671∗∗∗ 1.241∗∗∗
(0.414) (0.316)
Local dummy × Lockdown order dummy −0.614∗∗∗ −0.417∗∗∗
(0.091) (0.080)
Footprint dummy 0.684∗∗∗ 0.486∗∗
(0.248) (0.203)
Local dummy × Footprint dummy −0.704∗∗∗ −0.501∗∗∗
(0.099) (0.087)
Footprint change (%) −0.009∗∗∗ −0.005∗
(0.003) (0.002)
Local dummy × Footprint change (%) 0.011∗∗∗ 0.008∗∗∗
(0.001) (0.001)
Covid cases per population 45.505∗∗ 36.386∗∗
(18.359) (16.779)
Local dummy × Covid cases per population −149.913∗∗∗ −117.687∗∗∗
(18.915) (17.053)
Fund FE Yes Yes Yes Yes Yes Yes Yes Yes
Year-month FE Yes Yes Yes Yes Yes Yes Yes Yes
Observations 34,569 34,569 34,569 34,569 34,569 34,569 34,569 34,569
Adjusted R2 0.058 0.057 0.057 0.057 0.047 0.047 0.047 0.047
60

Table 2.4: Impact of lockdowns on stock-level investment returns
This table presents the results of the impact of lockdowns on stock-level investment returns.
Each fund’s monthly investment returns for every position are calculated using the dollar value
of holdings, assuming that positions are bought or sold at the previous month’s end price, and
adjusting for the initiation of new positions. Three-month investment returns are computed by
cumulating returns in a rolling window at the monthly frequency. I include fund-stock pairs if a
fund holds any position in a stock at least for a month during the sample period. For the months
with no holding for a pair, either because a fund made a new investment decision after January
2019 or because a fund has liquidated the position, the returns are set to 0%. Panel (a) presents
the summary statistics, and Panel (b) reports the results of stock-level regressions on investment
timings for different time periods:
Returni,j,t = β0 + β1Locali,j + β2P ostLockdownj,t + β3Locali,j × P ostLockdownj,t
+ F undF Ei + StockF Ej × P ostLockdownF Ej,t + ϵi,j,t
where Returni,j,t is three-month investment return of fund i on stock j in month t. Locali,j
is a dummy equal to 1 if the distance between fund i and firm j is less than 100 miles,
P ostLockdownj,t is a dummy indicating a post lockdown period based on the footprint in the
local area of firm j’s headquarters. Standard errors are clustered by stock, fund, and time.
(a) Summary statistics
Statistic N Mean Median St. Dev. Min Pctl(25) Pctl(75) Max
Monthly return (%) 3,144,411 0.57 0.00 8.60 −28.46 0.00 2.20 35.01
Three-month return (%) 2,805,633 1.64 0.00 14.13 −35.82 −0.66 5.37 56.82
(b) DiD regression result
Dependent variable:
Three-month return (%)
Until Sep 2020 Until Dec 2020
(1) (2) (3) (4)
Local −0.024 −0.041 −0.025 −0.027
(0.062) (0.039) (0.062) (0.036)
Local × Post-lockdown −0.052∗∗∗ −0.043∗∗ −0.038∗∗ −0.028∗
(0.010) (0.016) (0.015) (0.016)
Stock x Post-lockdown FE Yes Yes Yes Yes
Fund FE Yes Yes
Observations 2,805,633 2,805,633 3,281,594 3,281,594
Adjusted R2
0.144 0.146 0.112 0.115
61

Table 2.5: Impact of lockdowns on buy/sell decision, trade size, and turnover ratio
This table reports the regression results of the following specifications:
StockDGTWj,t+1 = β0 + β1Locali,j + β2P ostLockdowni,t + β3Locali,j
× P ostLockdowni,t + StockDGTWj,t
+ StockF Ej × F undF Ei + T imeF Et + ϵj,t
where StockDGTWj,t+1 is the signed next month’s DGTW return of stock j that is bought by
fund i in month t. The returns of sold stocks are multiplied by −1 so that the negative next-period
stock return indicates poor investment timings.
Column (1) presents the results only with positions with increased holdings. Column (2) compares
the results on buy- and sell-orders by including both the positions with increased and decreased
holdings, and interacting the DiD interaction term with a dummy Buy which is 1 for bought
stocks and 0 for sold stocks. Columns (3) and (4) show the results with the log of the dollar value
of trading amounts as the dependent variable. Standard errors are clustered by fund and stock.
Columns (5) and (6) present fund-level regression results using the monthly turnover ratio as the
dependent variable. Standard errors are clustered by fund.
Dependent variable:
DGTWt+1(%) ln(Trade size) Turnover ratio(%)
Buy Buy/Sell Buy Buy/Sell
(1) (2) (3) (4) (5) (6)
Local -0.857∗∗∗
(0.257)
Post-lockdown −0.002 −0.002 0.052 0.091 0.682 0.367∗
(0.003) (0.002) (0.108) (0.065) (0.444) (0.184)
Buy 0.001 0.310∗∗∗
(0.002) (0.040)
DGTWt −0.143∗∗∗ −0.085∗∗∗
(0.008) (0.008)
DollarInvestedt−1 0.000∗∗∗ 0.000
(0.000) (0.000)
Local × Post-lockdown −0.006∗
0.007∗∗ −0.052 0.050 −0.646
(0.004) (0.002) (0.048) (0.040) (0.370)
Local × Buy 0.002 0.027
(0.002) (0.045)
Post-lockdown × Buy 0.000 −0.192∗∗∗
(0.004) (0.053)
Local × Post-lockdown × Buy −0.012∗∗ −0.093
(0.005) (0.065)
Stock×Fund FE Yes Yes Yes Yes
Fund FE Yes Yes
Year-month FE Yes Yes Yes Yes Yes Yes
Observations 255,154 584,840 281,707 636,397 17,895 34,847
Adjusted R2
0.078 0.036 0.712 0.656 0.231 0.279
62

Table 2.6: Speed of Information Diffusion (SID) within fund families
Table 2.6 presents the results on the speed of information diffusion (SID) within fund families,
which is calculated following Cici, Jaspersen, and Kempf (2017) by taking the average of ID for
each stock initiation in a fund family:
IDf,s,q =
If,s,q − 1
If,s,q + Jf,s,q − 1
where If,s,q is the number of funds in family f that initiates a position in stock s that is not
already held by any fund in the family in quarter q, Jf,s,q is the number of funds in the family
that follow later during the information interval. The information interval starts when the initial
stock purchase happens and ends when at least one initiating manager liquidates.
Panel (a) presents the t-test results that compare the overall, within-style, and across-style SIDs
before and after lockdowns. Post-lockdown indicates the quarters starting from 2020 Q2. Panel
(b) presents a within-family change in SID. Standard errors are clustered by fund family. Panel
(c) presents the regression results of the stock-level regression on investment timings as in Equation (2.3), additionally interacting the interaction term with a dummy HighSID, which indicates
funds from families with above median SID during 2019. Panel (d) shows the results on investment timings when the dummy Local is replaced with a dummy Branch, which is 1 for the stocks
headquartered in states where a fund family branch exists, and 0 for the stocks headquartered in
states without a branch. Standard errors are clustered by fund and stock.
(a) T-tests of SID before and after lockdowns
Pre-lockdown Post-lockdown Difference t-statistic
SID 0.612 0.590 −0.022 −6.153
SIDwithin 0.618 0.604 −0.013 −2.865
SIDacross 0.609 0.578 −0.032 −9.864
(b) SID before and after lockdowns within fund families
Dependent variable:
SID SIDwithin SIDacross
(1) (2) (3)
Post-lockdown −0.021∗∗∗ −0.016∗ −0.029∗∗∗
(0.006) (0.008) (0.006)
Fund family FE Yes Yes Yes
Observations 5,789 4,302 5,560
Adjusted R2
0.389 0.450 0.395
63

(c) Heterogeneity of investment timings across funds with different SIDs
Dependent variable:
Three-month return (%)
SID SIDW ithin SIDAcross
(1) (2) (3) (4) (5) (6)
Local 0.120 0.019 0.069 0.037 0.061 −0.014
(0.098) (0.062) (0.123) (0.077) (0.097) (0.068)
High SID −0.037 0.087 −0.040
(0.066) (0.077) (0.063)
Local × Post-lockdown −0.203∗ −0.197∗ −0.246∗ −0.234∗ −0.142 −0.140
(0.116) (0.116) (0.141) (0.140) (0.123) (0.123)
Local × High SID −0.239∗∗ −0.069 −0.081 −0.053 −0.136 0.005
(0.113) (0.089) (0.140) (0.097) (0.118) (0.098)
Post-lockdown × High SID 0.051 0.039 0.056 0.047 −0.069 −0.069
(0.091) (0.091) (0.098) (0.096) (0.096) (0.096)
Local × Post-lockdown × High SID 0.202 0.199 0.183 0.172 0.060 0.063
(0.160) (0.161) (0.168) (0.168) (0.169) (0.170)
Stock x Post-lockdown FE Yes Yes Yes Yes Yes Yes
Fund FE Yes Yes Yes
Observations 2,302,382 2,302,382 2,148,940 2,148,940 2,270,571 2,270,571
Adjusted R2
0.145 0.147 0.144 0.146 0.145 0.147
(d) Investment timings on the stocks headquartered in the states with a fund family branch
Dependent variable:
Three-month return (%)
Until Sep 2020 Until Dec 2020
(1) (2) (3) (4)
Branch −0.028 −0.048 −0.029 −0.048
(0.054) (0.047) (0.054) (0.049)
Branch × Post-lockdown −0.090 −0.083 −0.041 −0.033
(0.080) (0.080) (0.079) (0.079)
Stock x Post-lockdown FE Yes Yes Yes Yes
Fund FE Yes Yes
Observations 2,734,890 2,734,890 3,199,268 3,199,268
Adjusted R2
0.144 0.146 0.112 0.115
64

Table 2.7: Public information seeking
Table 2.7 reports the results on fund managers’ reliance on public information (RP I) before and
after lockdowns, which is calculated following Kacperczyk and Seru (2007). I run the following
cross-sectional regression for each fund f and quarter q using all stocks s=1 to n in the fund’s
portfolio:
%∆Holdf,s,q =β0,q + β1,q∆Res,q−1 + β2,q∆Res,q−2 + β3,q∆Res,q−3+
β4,q∆Res,q−4 + ϵf,q, ∀s = 1 · · · n
where %∆Holdf,s,q denotes a percentage change in the number of holdings or dollar value of
holdings of stock s held by fund f from quarter q − 1 to q, ∆Res,q−p measures a change in the
recommendation of the consensus forecast of stock s from quarter q − p − 1 to quarter q − p,
and p = 1, 2, 3, 4 is the number of lags of the forecast. %∆Holdf,s,q is set to 100% when a new
stock position is initiated. Reliance on Public Information (RP I) is the unadjusted R2 of the
regression. I denote the measure as RP I if the number of shareholdings is used and as RP Idollar
if the dollar value of holdings is used to compute %∆Holdf,s,q.
Panel (a) presents the t-test results that compare RP I(RP Idollar) before and after lockdowns,
where post-lockdown refers to the quarters starting from 2020 Q2. Panel (b) presents fund-level
regression results using RP I calculated separately for local and non-local holdings as the dependent variable. Standard errors are clustered by fund.
(a) T-tests of RP I before and after lockdowns
Pre-lockdown Post-lockdown Difference t-statistic
RP I 0.074 0.077 0.002 1.378
RP Idollar 0.075 0.079 0.004 2.205
(b) Local and non-local RP I before and after lockdowns
Dependent variable:
RP I RP Idollar
(1) (2)
Local 0.054∗
0.054∗∗
(0.028) (0.021)
Post-lockdown 0.001 0.002
(0.002) (0.002)
Local × Post-lockdown 0.006 0.022∗∗
(0.013) (0.009)
Fund FE Yes Yes
Observations 6,091 6,091
Adjusted R2
0.235 0.248
65

Table 2.8: Impact of lockdowns on local bias
This table presents the regression results about the impact of lockdowns on local bias:
LocalBiasi,t = β0 + β1Lockdowni,t + β2Lockdowni,t × T2i + β3Lockdowni,t × T3i
+ F undF Ei + T imeF Et + ϵi,t
where the dependent variable is the local bias of fund i in month t. Lockdowni,t is defined in four different
ways: 1) a dummy variable equal to 1 if the zip code of a fund had a stay-at-home order implemented 2) a
dummy variable equal to 1 if the footprint activity dropped by more than 30% relative to the 2019 average
3) percentage change in footprint activities relative to the 2019 average 4) state-level Covid case counts
scaled by the number of the local populations. Lockdowni,t is interacted with a categorical variable that
divides funds into terciles based on their local bias in 2019. Standard errors are clustered by fund.
Dependent variable:
Local (100 miles) bias (%)
(1) (2) (3) (4)
Lockdown order dummy 0.533∗∗∗
(0.182)
Footprint dummy 0.476∗∗∗
(0.168)
Footprint change (%) −0.013∗∗∗
(0.004)
Covid cases per population 51.478∗∗∗
(21.914)
Lockdown order dummy × T2 −0.454∗∗∗
(0.173)
Lockdown order dummy × T3 −0.820∗∗∗
(0.200)
Footprint dummy × T2 −0.469∗∗∗
(0.175)
Footprint dummy × T3 −0.850∗∗∗
(0.198)
Footprint change (%) × T2 0.006∗∗
(0.003)
Footprint change (%) × T3 0.011∗∗∗
(0.003)
Covid cases per population × T2 −95.918∗∗∗
(30.745)
Covid cases per population × T3 −117.897∗∗∗
(33.997)
Fund FE Yes Yes Yes Yes
Year-month FE Yes Yes Yes Yes
Observations 16,879 16,879 16,879 16,879
Adjusted R2
0.954 0.954 0.954 0.953
66

Table 2.9: Social index
This table presents the regression results on the impact of lockdowns on the investment performance in Section 4 and the results on local bias in Section 6, separately for the funds in the
regions with high and low social index. A social index measure is obtained from the Northeast
Regional Center for Rural Development at the Pennsylvania State University. The high and low
social index refers to the highest and the lowest terciles.
(a) Impact of lockdowns on portfolio-level return
Dependent variable:
100 miles 30 miles
Benchmark excess DGTW Benchmark excess DGTW
Social index Low High Low High Low High Low High
(1) (2) (3) (4) (5) (6) (7) (8)
Local −0.382∗∗∗−0.313∗∗∗−0.272∗∗∗−0.115∗−0.422∗∗∗−0.379∗∗∗−0.338∗∗∗−0.148∗
(0.082) (0.065) (0.074) (0.060) (0.092) (0.083) (0.081) (0.076)
Post-lockdown 1.055∗∗∗ 0.742∗∗ 0.866∗∗∗ 0.299 1.276∗∗∗ 0.389 1.198∗∗∗ 0.145
(0.352) (0.326) (0.294) (0.279) (0.430) (0.338) (0.344) (0.285)
Local × Post-lockdown−0.175−1.277∗∗∗ −0.084−0.748∗∗∗ −0.241−0.910∗∗∗ −0.055−0.338∗
(0.199) (0.184) (0.171) (0.163) (0.237) (0.229) (0.202) (0.190)
Fund FE Yes Yes Yes Yes Yes Yes Yes Yes
Year-month FE Yes Yes Yes Yes Yes Yes Yes Yes
Observations 10,366 8,606 10,366 8,606 9,738 8,066 9,738 8,066
Adjusted R2
0.060 0.059 0.050 0.048 0.048 0.053 0.043 0.043
67

(b) Impact of lockdowns on local bias
Dependent variable:
Local bias (100 miles) Local bias (30 miles)
Social index Low High Low High
(1) (2) (3) (4)
Post-lockdown dummy 0.724∗∗∗ 0.377 0.359 0.196
(0.273) (0.325) (0.245) (0.261)
Post-lockdown dummy × T2 −0.717∗∗ −0.595∗ −0.173 −0.377
(0.307) (0.333) (0.258) (0.398)
Post-lockdown dummy × T3 −1.073∗∗∗ −1.389∗∗∗ −0.498∗ −1.381∗∗∗
(0.334) (0.378) (0.297) (0.323)
Fund FE Yes Yes Yes Yes
Year-month FE Yes Yes Yes Yes
Observations 5,055 4,183 4,430 3,625
Adjusted R2
0.959 0.961 0.958 0.953
68

Chapter 3
Collective Cognition of Mutual Fund Teams
3.1 Introduction
Over the past 30 years, team management has become a dominant form of decision-making unit
in the mutual fund industry. While the share of team management remained below 20% before
1990, the number has continuously increased to reach almost 80% in 2020 (Figure 3.1). Given the
prevalence of team management, several studies explore the superiority of team management
over solo management with a focus on the fund-level investment performance (Prather and Middleton 2002; Chen, Hong, Huang, and Kubik 2004; Bär, Niessen-Ruenzi, and Ruenzi 2007; Adams,
Nishikawa, and Rao 2018).
Given this evolution of the management structure, one aspect of mutual fund managers’ investment decision-making that is worth revisiting is the behavioral biases that they display. Prior
papers document that professional investors tend to buy overvalued stocks (Edelen, Ince, and
Kadlec 2016), sell gains more than losses (Cici 2012; Lu, Ray, and Teo 2016), trade too much
(Puetz and Ruenzi 2011), and tend to trade attention-grabbing stocks (Fang, Peress, and Zheng
69

2014; Hartzmark 2015; Ben-Rephael, Da, and Israelsen 2017; Schmidt 2019). However, prior studies that do not carefully take into consideration the entities making such decisions leave the
question of the impact of team management on such behavioral biases unanswered.
In this paper, I explore if team management mitigates or aggravates one of the biases that
are known to be driven by the limited cognitive ability of fund managers: Rank effect. First documented in Hartzmark (2015), mutual fund managers tend to sell extreme winning and losing
positions while ignoring the positions in the middle, when the stocks are ranked based on the
return from the purchase price. The effect is attributed to the use of heuristic simplification in intuitive judgments under System 1 of the dual cognitive theory.∗ Because the cognitive processing
power is scarce and extreme ranks are salient (Diecidue and Wakker 2001), managers subconsciously include extreme-ranked positions into the consideration set for selling. By focusing on
a bias that is likely to be caused by cognitive constraints rather than by behavioral preferences,
I test if the collective cognitive ability of mutual fund teams reduces the use of heuristics when
making investment decisions.
The idea of collective cognition in psychology literature, defined as the group process involved
in the acquisition, storage, transmission, manipulation, and use of information (Von Cranach,
Ochsenbein, and Valach 1986; Wegner 1987; Hinsz, Tindale, and Vollrath 1997), suggests that
groups can be more cognitively sophisticated than individuals. Interpersonal interaction among
team members is one of the elements that enhance the cognitive performance of teams. Studies
document that a higher amount of spoken or written communication in face-to-face or online
∗See Simon (1956), Tversky and Kahneman (1974), Kahneman, Slovic, Slovic, and Tversky (1982), Epstein (1994),
Sloman (1996), Kahneman (2003), and Kahneman (2011) for the discussion on the dual cognitive theory.
70

groups leads to superior performance (Engel, Woolley, Jing, Chabris, and Malone 2014), particularly when communication and work contribution are equally distributed among group members
(Woolley, Chabris, Pentland, Hashmi, and Malone 2010; Engel, Woolley, Jing, Chabris, and Malone 2014; Kim, Engel, Woolley, Lin, McArthur, and Malone 2015).
At the same time, other reasons make the improvement in team decision making unclear.
Under groupthink, team members tend to conform to the predominant values or actions under
strong social pressure for unity (Janis 1982; Bénabou 2013). Under the pressure, members avoid
being harsh in critically evaluating colleagues’ ideas, which may induce rather than curtail the
use of heuristics in making investment decisions. In this case, the interactions within a cohesive
group may aggravate rather than alleviate behavioral biases of fund managers.
Considering the potential for both positive and negative impacts of team management on fund
managers’ behavioral biases, I document that the rank effect, which is based on the objective ranking of positions in portfolios, is less susceptible to groupthink. Other than heuristic simplification
from which the rank effect is suggested to be arising, two other psychological elements that explain the biases in behavioral finance are affective short-circuiting and self-deception (Hirshleifer
2015). The subjectiveness embedded in the two elements, feelings, and self-esteem, may aggravate the biases in teams. Moods including excitement, anxiety, or fear are easy to be transmitted
or amplified through social interactions (Quinn and Dutton 2005; Parkinson and Simons 2009).
Furthermore, the biases based on a desire to maintain self-esteem are vulnerable in a social context as people dislike damaging their reputation (Heimer 2016; Dorn and Yadav 2022). Relative to
these two elements, I suggest that the errors arising from heuristic simplification including the
rank effect are more likely to be detected and corrected through the discussions in teams, as the
same set of extreme-ranked stocks is salient to all team members.
71

The mutual fund industry is a useful laboratory to test whether a team of professional investors displays superior cognitive ability than individuals. This is because the percentage of
team-managed funds has been continuously increasing, and a significant share of fund managers
have both experiences in solo and team management. As the rank effect of the same fund manager is observed both under solo and team management, I can measure the effect of collective
cognition in teams that is distinct from the effect of individual managers’ characteristics.
I first replicate Hartzmark (2015) with the sample period from January 1980 to December 2021
to show the existence of the rank effect in the recent period. Stock×Time fixed effects are included
to compare a specific stock’s probability of being sold by two funds at the same time period, one
in which the stock is extreme-ranked and another in which the stock is not extreme-ranked. The
fixed effects isolate the effect of the saliency measured by extreme ranks from the effect of firmspecific information or the level of returns itself. I find that the probability of being sold increases
by 9.7% when a stock is best ranked and by 8% when it is worst ranked relative to when it is not
extreme-ranked, after controlling for the holding period. The numbers are comparable to 6.8%
and 9.7% documented in Hartzmark (2015).
Additionally, I show that the rank effect is value-destroying by comparing the post-trade
returns of extreme-ranked stocks to those of non-extreme-ranked stocks and random stocks in
a portfolio that are sold. Specifically, three-month post-trade returns of the sold stocks with
extreme ranks are lower by 18bps than those of the sold stocks with non-extreme ranks, and by
16bps points than the random selling strategy on average. The result supports the idea that fund
managers’ tendency to sell extreme-ranked positions is not the result of their rational investment
strategies but the result of employing heuristic simplification.
72

Next, I explore how the magnitude of the rank effect is different under solo versus team management. I limit the sample to the stocks that are extreme-ranked in at least one sample fund in
a given quarter, and run similar regressions as before but interact extreme rank dummies with
a team dummy indicating whether a fund is team-managed or not. Fund family fixed effects are
included to control for the different team operating styles in different organizations. I find that
in teams, the probability of selling extreme-ranked stocks drops by 5% for the best and 20% drop
for the worst-ranked stocks. When manager fixed effects are included to examine the withinmanager effect, the magnitude of the changes drop by half but the direction remains consistent.
The result provides preliminary evidence that teams are less likely to display the rank effect than
individuals.
The concern with the previous results is that solo and team management are not randomly
assigned to funds. Without random assignment, the effect may be driven by omitted variables that
are correlated with team assignments and the tendency to use heuristics. To address this issue, I
perform propensity score matching to identify the counterfactuals of 342 funds that switch from
solo to team management. With the matched sample, I run triple difference-in-differences to
identify the change in the probability of selling extreme-ranked stocks relative to that of middleranked stocks when a fund switches from solo to team management. Consistent with my previous
findings, switching funds exhibit a larger decrease in the probability of selling extreme-ranked
positions relative to middle positions. On average, this decrease is 6 times greater for the bestranked positions and 11 times greater for the worst-ranked positions compared to non-switching
funds.
Next, I explore if the inter-group discussion is the mechanism through which teams reduce
the rank effect, by exploring if the reduction of the rank effect is greater for the extreme-ranked
73

positions that are more likely to be discussed in teams. As small-group members tend to focus
their discussion on the information they have in common (Wittenbaum and Stasser 1996; Stasser
and Birchmeier 2003; Wittenbaum, Hollingshead, and Botero 2004), I posit that managers spend
more time discussing extreme-ranked stocks that are also extreme-ranked in other funds that
team members manage at the same time. I find that the rank effect on positions that are extremeranked in multiple funds becomes close to zero or statistically insignificant. This suggests that
group discussions enable team members to engage System 2 to reduce the use of heuristics when
making decisions to sell.
Finally, I test the value of cognitive diversity by exploring whether the rank effect is more
likely to decrease in teams with members who exhibit varying susceptibility to the rank effect. I
compute the rank effect at the individual level to categorize mutual fund teams into cognitively
heterogeneous and homogeneous teams. By comparing the rank effect observed in teams with
the average rank effect displayed by team members under solo management, I find that the beneficial effect of team discussion in reducing the rank effect is more pronounced in cognitively heterogeneous teams. These results suggest that collective cognition is more effective when team
members have diverse cognitive styles, enabling team discussions to better detect and correct
errors.
The main contribution of this paper is to provide evidence of the superior cognitive ability of
mutual fund teams. While prior papers document the behavioral bias of professional investors
arising from limited attention (Fang, Peress, and Zheng 2014; Hartzmark 2015; Ben-Rephael, Da,
and Israelsen 2017; Schmidt 2019), they do not touch on whether a group of professional investors
74

can pool their attention to reduce the bias. By documenting the beneficial role of teams in reducing bias based on heuristic simplification, I provide new evidence on the role of teams in asset
management.
This adds to the recent studies on the impact of team management on behavioral bias. For
instance, Barahona, Cassella, and Jansen (2022) documents a beneficial effect of teams in reducing
return extrapolation bias, while Dorn and Yadav (2022) reports a detrimental effect of teams by
showing a magnified disposition effect in teams. Additionally, Antoniou and Mitali (2018) find
that teams irrationally make decisions based on positive experiences due to the supportive signals
from team members. By focusing on the interactive nature of team members to suggest collective
cognition as another benefit of team portfolio management, I contribute to this new strain of
contemporary research.
3.2 Data
This study requires data of three types: (1) Quarterly mutual fund holdings (2) Mutual fund manager and team information (3) Mutual fund returns and characteristics.
Focusing on US active equity mutual funds from January 1980 to December 2021, I obtain
fund holdings from Thomson Reuters for 1980-2007 and Survivor-Bias-Free US Mutual Funds
(CRSP MF) for 2008-2021, fund manager and team information from Morningstar Direct, and
fund returns and characteristics from CRSP monthly stock file and CRSP MF.
I obtain the sample of interest starting from funds with non-missing monthly returns and
TNA from CRSP MF. The data is then merged with the fund information from Morningstar, which
includes domestic non-index equity funds with monthly returns and total net asset value. As the
75

two databases have different fund identifiers at the share class level, CRSP_F UNDNO for
CRSP MF and SecId for Morningstar, I perform the merge following prior studies as follows.
First, the tickers and cusips in CRSP MF are forward- and backward-filled. Following Pástor,
Stambaugh, and Taylor (2015), the latest identifiers are selected if they are time-varying, and set
to missing values if they correspond to more than one share class, CRSP_F UNDNO. The
tickers and cusips in Morningstar are cleaned the same way but using SecId for the share class
identifier. Following Berk and Van Binsbergen (2015), missing ticker and cusip in CRSP MF and
are replaced with CRSP_F UNDNO and those in Morningstar are replaced with SecId to make
sure that missing values are not matched.
Second, CRSP MF and Morningstar data are merged using the cleaned tickers and cusips. Following Barahona, Cassella, and Jansen (2022), the remaining non-merged funds are then merged
based on an exact match between year, month, monthly return, and monthly total net asset value.
The remaining non-matched funds are then matched exactly based on year, month, total net asset
value, and nearly based on monthly returns with differences of at most two basis points. Finally,
the remaining non-matched funds are matched exactly based on year, month, monthly return,
and nearly based on total asset value with differences of at most $20,000.
Third, to correct the possible errors in the merging process, I follow Berk and Van Binsbergen
(2015) to drop the funds that the same share classes are matched less than 60% of the time. The
matches with a rate higher than 60% are considered as the correct match. Following Pástor,
Stambaugh, and Taylor (2015), the fund observations that do not have all of the share classes
matched based on Morningstar fund level identifier F undID are dropped.
Forth, the CRSP fund share classes that are domestic equity but not index is selected using
crsp objective code, strategic insight objective code, index fund flag, and fund names. Following
76

Hartzmark (2015), only the portfolios with the number of stock holdings greater than 20 and less
than 500 are included in the sample. To calculate the rank of stocks within a portfolio using
returns from purchase, positions with unknown purchase prices are dropped.
Fifth, quarterly fund holdings are matched to the funds that are selected in previous steps.
Considering that Thomson Reuters’s coverage deteriorates in 2008 and excludes many young
funds (Zhu 2020), quarterly holdings from Thomson Reuters are used from 1980 to 2007. I use
MFLINKS dataset to match the Thomson Reuters identifier W F ICN to CRSP identifier
CRSP_F UNDNO, and keep distinct portfolios. Quarterly holdings from 2008 to 2021 are obtained from CRSP MF.
Finally, stock price information obtained from CRSP is merged with the data to adjust the
quarterly fund holding information. Stocks with share codes 10 and 11 are included, and the
observations with missing price information are dropped. Fund manager and mutual fund team
information obtained from Morningstar is merged into the dataset. The observation of the data
is at the fund manager-fund-stock-quarter level.
Table 1 presents the summary statistics. There are 2,929 unique funds with the fund manager
or team information and 7,258 unique fund managers. 67% of sample funds are team-managed.
The median fund total net assets (TNA) is $145 million, the median fund age is 2.5 years, and the
median number of stock holdings is 77. On average, funds are managed by 2.84 managers with
10.56 years of tenure.
77

3.3 Rank effect
In this section, I replicate Hartzmark (2015) for the sample period from January 1980 to December
2021 to explore the degree of rank effect in a longer sample period including recent years. Posttrade returns of traded stocks are examined to measure the performance of the stocks that are
traded due to their extreme ranks.
3.3.1 Rank effect
Hartzmark (2015) documents that individuals and mutual fund managers tend to sell the best and
worst ranked stocks when positions are ranked by the return from purchase, with the sample
period from 1990 to 2010. The effect is suggested to be arising from the saliency of extremeranked stocks and investors’ limited attention, under which investors subconsciously include
extreme-ranked stocks into a consideration set for selling.
Psychology and behavioral economics literature suggest that human cognitive processes are
divided into two parts, System 1 and System 2 (Sloman 1996; Stanovich and West 2000; Kahneman and Frederick 2004; Kahneman 2011). In the dual process theory of cognition, System 1 is
characterized as automatic, rapid, and effortless while System 2 is characterized as deliberate,
slow, and deliberate.
The intuitive intelligence under System 1 can be useful when cognitive resources are limited,
particularly if the individual making decisions is equipped with the required skill and proficiency.
However, judgments made in System 1 are often susceptible to errors as the decision-making
process is spontaneous and indeliberate. The shortcuts or heuristic rules people follow under
System 1 often have limited validity, and could lead to systematic errors and biases (Tversky and
78

Kahneman 1974). The rank effect could be understood as the errors made in the process of fast
judgments under System I, in which investors neglect non-salient information and tend to make
decisions prioritizing salient, extreme-ranked stocks in their portfolios.
A major concern with using portfolio ranks to proxy for the saliency of a stock is that the
rank of a stock can be correlated with firm-specific information and the level of returns itself.
To address this issue, the rank effect is measured by comparing two fund managers’ different
tendencies to sell the same stock, but with different ranks within each of their portfolios. As
the rank is not a stock-specific but a portfolio-specific measure of saliency, their different selling
decisions on the stock can be attributed to the effect of the portfolio-specific ranks rather than
the stock-specific attributes. Specifically, I estimate the following linear probability model as in
Hartzmark (2015):
Selli,j,t = β0 + β1Besti,j,t + β2W orsti,j,t + StockF Ei × QuarterF Et + ϵi,j,t (3.1)
where Selli,j,t is a dummy equal to 1 if fund j liquidates (all shares are sold) or sells (a decrease in the number of shares held) stock i in quarter t, Besti,j,t (W orsti,j,t) is a dummy variable
equal to 1 if stock i has the highest (lowest) rank in the portfolio of fund j in quarter t when
stocks are ranked based on their return from the purchase price. Returns at the time of sale are
calculated between the purchase price and selling price based on the report dates in the quarterly holding information. For the stocks that additional shares are purchased after the initiation,
the value-weighted average price is used to calculate the purchase price. Stock×Quarter fixed
effects control for any time-varying stock characteristics. The fixed effects allow the comparison
79

of selling decisions based on the variation of the ranks of a specific stock across fund managers’
portfolios. Standard errors are clustered by fund and time.
Table 3.2 Column (1) presents the regression results. Overall, the statistically significantly
positive coefficients on the extreme dummy variables indicate that the same stock becomes more
likely to be sold if the stock is extreme-ranked in a portfolio. Specifically, Column (1) suggests
that fund managers are 8.1% more likely to sell best-ranked stocks and are 5.7% more likely to
sell worst-ranked stocks.
To take into account that the length of time a position has been held may influence the rank
of a stock and fund managers’ selling decisions, Stock×time×holding period fixed effects are introduced in Column (2). By additionally matching positions with holding periods, the regression
examines the selling decisions of two funds that have held a specific stock for the sample length
of time, but the stock is extreme-ranked in one fund but not in another. The results indicate that
fund managers are 9.7% more likely to sell the best and 8% more likely to sell the worst position.
The numbers are comparable to 6.8% and 9.7% documented in Hartzmark (2015).
Column (3) presents the result when Stock×time×initiation quarter fixed effects are introduced, in which positions are matched based on the date the positions were initially added to a
portfolio. Although the magnitude of coefficients becomes smaller, the results remain consistent
with those of other specifications.
Together, the results confirm the existence of the rank effect suggested in Hartzmark (2015)
in the longer sample period. They suggest fund managers’ continued reliance on heuristic simplification, extreme ranks in this case, when making decisions to sell.
80

3.3.2 Post-trade returns
Even though the result in the previous section confirms mutual fund managers’ tendency to trade
salient, extreme-ranked stocks, the result alone is not sufficient to argue that fund managers are
making mistakes due to their limited cognitive resources. If selling extreme-ranked stocks generates high investment returns, the rank effect would reflect a deliberately designed trading strategy
rather than fund managers employing heuristic simplification due to their limited attention.
To examine whether the rank effect leads to value-destroying trades, I compare the post-trade
returns of the extreme-ranked stocks that are sold to the post-trade returns of the non-extremeranked stocks that are sold. If the selling decisions on extreme-ranked stocks are the result of
indeliberate judgments relative to those on non-extreme ranked stocks, I expect the post-trade
returns of extreme-ranked stocks to be smaller than those of non-extreme stocks that are sold.
To calculate factor-neutral stock returns, I first estimate monthly alphas of each stock in a
portfolio using Fama-French/Carhart four factors (Fama and French 1993; Carhart 1997). I use
a rolling window of the previous sixty months to estimate each stock’s factor exposures. I then
use these exposures to estimate the stock’s risk adjusted returns on the following month by subtracting the portion of stock returns that are the result of factor exposures from the stock returns.
Specifically, I obtain monthly factor-neutral stock return by subtracting the inner product of factor loadings and factor realizations from excess return for each stock-month as follows:
R
F N
i,t ≡ Ri,t − Rf − Λ
′
i,t−1Ft
(3.2)
where Ri,t−Rf is stock i’s excess return on month t, Λ
′
i,t−1
is a (4×1) vector of estimated factor
loadings, and and Ft
is a (4×1) factor realizations. Under the assumption that factor loadings are
81

estimated correctly and are stable, RF N
i,t captures a stock return with zero exposure to the priced
risk factors at each month. The monthly returns are then cumulated over the next three, six, nine,
and twelve months after the quarter stocks are sold.
I compute RExtremeSell by taking the average of the factor-neutral returns of the stocks that are
extreme-ranked and sold, and RN onExtremeSell by taking the average of the factor-neutral returns
of the stocks that are non-extreme-ranked and sold. -1 is multiplied to the returns of the sold
stocks so that positive values indicate that trades help portfolio performance. If fund managers’
selling decisions based on the rank effect is value-destroying relative to those not based on the
rank effect, I expect the signed RExtremeSell to be smaller than RN onExtremeSell.
Figure 3.2 (a) presents the average RExtremeSell−RN onExtremeSell with 95% confidence intervals
calculated using heteroskedasticity and autocorrelation consistent standard errors. The negative
values in all time horizons indicate that the signed post-traded returns of the sold stocks that
are extreme-ranked are lower than those of non-extreme-ranked stocks on average, meaning
that fund managers should have held the extreme-ranked stocks relative to non-extreme-ranked
stocks. The economic magnitudes of the differences are significant, which are 18, 36, 52, and
54 bps at the 3, 6, 8, and 12 months horizons after the quarter stocks were sold. The results
suggest that selling decisions on extreme-ranked positions are value-destroying relative to those
on non-extreme-ranked positions.
Following Akepanidtaworn, Di Mascio, Imas, and Schmidt (2021), I also compare the trades
based on the rank effect to those of a random selling strategy. I compute the expected payoff of
the random selling strategy by taking the equal-weighted average of the realized factor-neutral
returns across stocks held in the portfolio, which is denoted as RHold.
82

Figure 2 (b) presents the average RExtremeSell − RHold with 95% confidence intervals. The
figure resembles Figure 2 (a), but the numbers are smaller in magnitude. The consistent results indicate that fund managers’ selling decisions on extreme-ranked stocks underperform a
simple random selling strategy. The results suggest that fund managers’ suboptimal decisions
to sell extreme-ranked stocks partly explain the overall value-destroying selling decisions that
Akepanidtaworn, Di Mascio, Imas, and Schmidt (2021) documents.
Together, the results suggest that the rank effect leads to value-destroying trades, possibly due
to the saliency of extreme ranks that grabs the attention of fund managers with limited cognitive
resources.
3.4 Rank effect in teams
This section explores how the rank effect arises differently in team versus solo management.
After providing suggestive evidence on the effect of teams, I perform propensity score matching
to address the concerns with non-random team assignments to funds.
3.4.1 Team dummy
To examine how the rank effect of team-managed funds is different from that of solo-managed
funds, I interact the extreme dummies in Equation (1), best and worst dummies, with the team
dummy indicating if a fund is team-managed or not in the following specification:
Selli,j,t = β0 + β1Extremei,j,t + β2T eamj,t + β3Extremei,j,t × T eamj,t
+ Numholdingsj,t + α
F E + ϵi,j,t
(3.3)
83

where Selli,j,t is a dummy equal to 1 if fund j liquidates (all shares are sold) or sells (a decrease
in the number of shares held) stock i in quarter t, Extremei,j,t is a dummy indicating if stock i
is best (worst) ranked in fund j in quarter t, and T eamj,t is a dummy equal to 1 if fund j is teammanaged and 0 if solo-managed in quarter t. Numholdingsj,t is the number of stock holdings of
fund j in quarter t, which controls for the effect of the number of stock holdings on the amount
of attention paid to extreme-ranked stocks and thus on the likelihood of stock sales. Standard
errors are clustered by fund and time.
Table 3.3 presents the regression results when observations are limited to the stocks that are
extreme-ranked in at least one fund in the sample. Columns (1), (3), and (5) present results similar
to those in Table 3.2 with the same set of fixed effects, and now the fund family fixed effects are
included to measure the within-fund family effect. The fund family fixed effects control for any
time-invariant different team operation styles in different fund management companies.
The coefficients of interest are those of interaction terms. The negative coefficients indicate
that fund managers’ tendency to sell extreme-ranked stocks is smaller in team-managed funds
relative to solo-managed funds. When the holding period is controlled for in Column (3), the
total effects indicate that team-managed funds’ probability to sell the best (worst) ranked stock
is smaller by 0.003 (0.008) percentage points than solo-managed funds on average, which corresponds to a 5% (20%) drop.
The concern with the result is that the result might be confounded by a manager effect, as
different managers may have a different tendency to pay attention to and trade extreme-ranked
stocks. To separate out the effect of teams from that of managers, I include manager fixed effects
instead of the fund family fixed effects to explore the within-manager variation. The manager
84

fixed effects allow the comparison of the rank effect of the same manager inside and outside of a
team.
Table 3.3 Columns (2), (4), and (6) present the results with manager fixed effects and the same
set of fixed effects as before. The negative coefficients on the interaction terms indicate that
managers are less likely to sell extreme-ranked stocks in teams. Specifically, when the holding
period is controlled for in Column (4), the total effects indicate that a fund manager’s probability
of selling the best (worst) ranked stocks relative to middle-ranked stocks decreases by 0.002 (0.003)
percentage points, which correspond to 3% (5.8%) drop. The smaller magnitudes than the results
with fund family fixed effects indicate that the decrease in rank effect in the previous results at
the fund level is partly attributed to the manager effect. Still, the results with consistent signs
suggest the existence of the team effect in reducing the rank effect.
Taken together, the results provide suggestive evidence that the collective cognition of mutual
fund teams reduces the use of heuristics and the rank effect in selling decisions.
3.4.2 Propensity score matching
A remaining concern with the previous results is that funds are not randomly assigned to be
team managed. In an ideal experiment, I would randomly assign some funds to team management and others to solo management. With perfect random assignment, the differences in the
rank effect of the two groups would be attributed to the effect of teams. However, without such
randomness, the result may be driven by omitted variables. If the reason a fund is operated in
teams is correlated with the fund managers’ tendency to use heuristics, the causal effect of teams
cannot be properly identified. For instance, if the fund managers of poorly performing funds are
85

susceptible to behavioral bias and if such funds are more likely to be managed by teams, it would
be important to control the investment performance of funds.
To address the concern, I perform propensity score matching using fund characteristics. I
identify 342 funds that switch from solo to team management and with a team composition lasting
for more than four consecutive quarters. For each switching fund (treatment group), I identify a
control fund with similar characteristics in the four quarters before the even quarter. To calculate
the propensity score, I use the average quarterly values of TNA, turnover ratio, fund flow, fund
flow volatility, four-factor alpha, and the four-factor alpha volatility. Each fund in the treatment
group is matched to the nearest neighbor, which is the fund with the closest propensity score in
the same period.
With the matched sample, I run triple difference-in-differences to identify the change in the
probability of selling extreme-ranked stocks relative to that of middle-ranked stocks when a fund
switches from solo to team management in the following regression specification:
Selli,j,t = β0 + β1Extremei,j,t + β2T eamj,t + β3T reatedj,t
+ β4Extremei,j,t × T eamj,t + β5Extremei,j,t × T reatedj,t
+ β6T eamj,t × T reatedj,t + β7Extremei,j,t × T eamj,t × T reatedj,t
+ Numholdingsj,t + α
F E + ϵi,j,t
(3.4)
where the definitions of the variables are the same as those of Equation (3). Standard errors are
clustered by fund and time. If teams alleviate fund managers’ tendency to resort to extreme ranks
when making selling decisions, I expect the drop in the probability of selling extreme positions
relative to middle positions to be bigger for the treatment group compared to the control group.
86

Table 3.4 presents the regression result with the same set of fixed effects as before. Overall,
the significantly negative coefficients on the triple interaction terms indicate that compared to
non-switching funds, the switching funds become less likely to sell extreme positions after the
switch. Figure 3.3 presents the results of Table 4 Column (4) with 95% confidence intervals, the
difference between the linear combination of regression coefficients separately for the middleranked, best-ranked, and worst-ranked positions. Figure 3 (a) shows that the DiD value of nonextreme positions is not statistically different from zero, indicating that the treatment and control
groups do not differ in the change of tendency to sell middle-ranked stocks before and after
treatment. On the other hand, the DiD value for the best-ranked position in Figure 3 (b) and the
DiD value for the worst-ranked position in Figure 3 (c) are negative and statistically significant.
The difference-in-difference values of the three groups yield statistically significantly negative
triple differences in Figure 3.3 (b) and (c). Relative to non-extreme positions, the drop in the DiD
value for the best positions is bigger by 0.075 percentage points and that of the worst position is
bigger by 0.13 percentage points, which corresponds to about six and eleven times bigger effects
in magnitude, respectively.
Together, the results suggest that fund managers’ tendency to sell salient positions decreases
when the decisions are jointly made in teams, although the results cannot rule out the effect of
unobservables.
3.5 Channel analysis
In this section, I explore the channel that enables teams to reduce the use of heuristics when
making decisions to sell. I explore the team discussion channel by examining the reduction in the
87

rank effect of the stocks that team members are more likely to discuss. Next, I explore if teams
with cognitive style diversity more effectively detect and correct the errors arising from the use
of heuristics.
3.5.1 Team member’s common holdings
In the dual theory of cognition, the slow judgments in System 2 monitor and revise the errors
arising in System 1. As Kahneman (2000) suggests, System 2 could be evoked if 1) decisionmaking individuals possess the required knowledge or 2) a relevant cue to evoke System 2 is
present. Under the assumption that fund managers operating funds in teams possess the required
knowledge to reduce the bias arising from saliency, I suggest that the team decision-making
process generates a cue to evoke System 2.
Specifically, I suggest the inter-group interactions fund managers have in teams as the process generating the cue for System 2. As small-group members tend to focus their discussion on
the information they have in common rather than what each of them uniquely possesses (Wittenbaum and Stasser 1996; Stasser and Birchmeier 2003; Wittenbaum, Hollingshead, and Botero
2004), I examine if the rank effect arises smaller for the stocks that are commonly extreme-ranked
in multiple funds.
To test the idea, I exploit the fact that fund managers often manage more than one fund at
the same time. Given that fund managers tend to discuss common information, I expect fund
managers in teams to devote more time and energy to discussing the stocks that are commonly
held and extreme-ranked in multiple funds managed by team members, which would invoke the
cue to make deliberate judgments under System 2. Specifically, I hypothesize that the rank effect
88

would arise smaller for the stocks that are extreme-ranked in another fund that team members
manage at the same time relative to the stocks that are extreme-ranked only in one fund.
To test the effect of the amount of team discussion on the magnitude of the rank effect, I
include only the funds that are team-managed and at least one of the team members manages
other funds at the same time. Also, I restrict the observations to the stocks that are also held by
at least one of the team members’ other funds. I run the following regression:
Selli,j,t = β0 + β1Besti,j,t + β2CommonBesti,j,t+
β3W orsti,j,t + β4CommonW orsti,j,t + Cj,t + α
F E + ϵi,j,t
(3.5)
Selli,j,t is a dummy equal to 1 if fund j liquidates (all shares are sold) or sells (a decrease in the
number of shares held) stock i in quarter t, Besti,j,t (W orsti,j,t) is a dummy indicating if stock i is
best (worst) ranked in fund j in quarter t. CommonBesti,j,t (CommonW orsti,j,t) is 1 if stock i is
also extreme ranked in other funds that fund j’s fund managers manage in quarter t. For control
variables, Cj,t, the number of holdings of fund j in quarter t and the number of other funds that
team members of fund j manage in quarter t are included. Standard errors are clustered by fund
and time.
The coefficients of interest are those of extreme and common extreme dummies. If the rank
effect arises smaller for the stocks that are extreme-ranked in multiple funds relative to those
extreme-ranked only in one fund, I expect the magnitude of the coefficients of common extreme
dummies to be smaller than those of extreme dummies, or statistically insignificant.
Table 3.5 presents the regression results with the same set of fixed effects as in the previous tests. In all specifications, the coefficients on the common extreme dummies are smaller
in magnitude than on extreme dummies and are statistically insignificant in some specifications.
89

Specifically, Column (4) with fund family and Stock×Time×Holding period fixed effects presents
that the coefficient on the best dummy is 0.066 while that of the common best dummy is 0.003
and is statistically insignificant. For the worst-ranked stocks, the coefficient on the worst dummy
is 0.061 while that on the common worst dummy is -0.007. The differences between the extreme
and common extreme coefficients are significant at the 95% confidence level.
The results show that the rank effect arises smaller or even disappears for the stocks that
are extreme-ranked in multiple funds that team members manage at the same time, relative to
the stocks that are extreme-ranked only in one fund. This suggests that team discussions induce
cues to operate System 2, which enables teams to reduce their use of heuristics based on salient,
extreme ranks when making decisions to sell.
3.5.2 Cognitive style heterogeneity
In this section, I examine if the heterogeneity of team members in their cognitive styles facilitates the reduction of rank effect in teams. Cognitive style refers to the way in which individuals
encode, process, and communicate information (Ausburn and Ausburn 1978; Kozhevnikov, Kosslyn, and Shephard 2005). Along with several studies that document the positive impact of the
demographic diversity of team members on the cognitive performance of teams, team members’
cognitive style diversity is suggested to positively affect the collective cognition of teams (Aggarwal, Woolley, Chabris, and Malone 2019). Aggarwal, Woolley, Chabris, and Malone (2019)
documents that a moderate amount of cognitive style diversity enhances the collective intelligence of teams when members with different information processing styles can communicate
effectively.
90

Following this line of research, I hypothesize that mutual fund teams with members of diverse
cognitive styles will be better at correcting errors arising from the fast judgments made in System
I, including the rank effect. Although Aggarwal, Woolley, Chabris, and Malone (2019) documents
the detrimental effect of excess cognitive style diversity due to the high cost of communication,
I expect the different cognitive styles of mutual fund team members to facilitate the reduction of
the rank effect as the median fund in the sample is managed by two fund managers.
I use each fund manager’s rank effect measured from their trades during solo management to
proxy for each manager’s cognitive style when making decisions to sell. Estimating the degree
of rank effect at the individual level allows me to determine the diversity of the teams that each
fund manager joins. The different team effects between homogenous and heterogenous teams
would indicate how the cognitive diversity of team members functions to reduce the rank effect
in teams.
Because every team member’s individual rank effect should be measured to determine the
cognitive diversity of teams, I limit the analysis to 643 sample fund managers who have both
experiences of solo and team management in their management history, and to the teams that
they join whose all team members have managed funds solo at some point in their careers, either
before, after, or at the same time they experience team management. To measure fund managers’
individual rank effect, the observations are pooled over the entire history of solo management
for each fund manager.
Unlike the previous regression specifications, I am no longer able to include Stock×Time
fixed effects because the regression is now run on each fund manager and not across different
fund managers. Therefore, I include a set of variables that could affect stock ranks and thus fund
managers’ selling decisions following Hartzmark (2015). As Ben-David and Hirshleifer (2012)
91

documents that investors become more likely to sell positions as the magnitude of gains or losses
become bigger, the level of a return from the purchase price, and whether a position is at a gain
or loss are controlled for. Also, the stock holding period and stock return volatility are included
in the regression as the stocks held for a longer period of time and those with high volatilities
are more likely to have extreme returns and thus more likely to be extreme-ranked in a portfolio.
Specifically, I run the following logit regressions on the pooled observation of each fund manager
j = 1... J, where J = 643:
Selli,t = β
j
0 + β
j
BestExtremei,t + β
j
W orstExtremei,t+
β
j
2
(Returni,t × Gaini,t) + β
j
3
(Returni,t × Lossi,t)+
β
j
4Gaini,t + β
j
5
(Returni,t × Gaini,t ×
p
HoldingP eriodi,t)+
β
j
6
(Returni,t × Lossi,t ×
p
HoldingP eriodi,t) + β
j
7
p
HoldingP eriodi,t+
β
j
8
(Gaini,t × V ariancei,t) + β
j
9
(Lossi,t × V ariancei,t) + ϵ
j
i,t
(3.6)
where Selli,t is a dummy variable equal to 1 if fund manager j sells stock i in quarter t, Besti,t
(W orsti,t) is a dummy variable equal to 1 if stock i is best (worst) ranked in fund j in quarter t.
Returni,t is a return from purchase and Gaini,t (Lossi,t) is a dummy variable equal to 1 if stock
i’s return from the purchase price is positive (negative) for fund manager j. Return is interacted
with Gain and Loss separately to allow the probability of sale to increase with the magnitude of
returns and to capture different slopes in the positive and negative domains. HoldingP eriod is
the number of months a position is held from the initial purchase quarter, and is interacted with
92

the Return × Gain and Return × Loss respectively. V ariance is the variance of monthly stock
prices during the previous 12 months from quarter t.
The coefficients of interest are β
j
Best and β
j
W orst. A positive beta with a large magnitude indicates that a fund manager j is more likely to be influenced by extreme ranks when making
decisions to sell, when the effect of absolute return levels, holding periods, and stock price variance on ranks are controlled for. Betas close to zero mean that the fund manager’s selling decision
is not influenced by the extreme rank of stocks, and negative betas indicate that the fund manager
is less likely to sell extreme-ranked stocks, which is the opposite of the rank effect.
To measure the effect of collective cognition teams that reduces the rank effect, I construct a
statistical team counterfactual, β
CF , by taking an average of the team members’ individual betas.
The beta of the statistical team serves as a counterfactual indicating how the rank effect would
arise in teams if there were no team effects. I then run paired t-tests to compare the betas of
statistical teams to those of actual teams. The betas of actual teams are computed in the same
way as in Equation (6), but now the regression is run on the observations of each team. If a team
with the same team member composition manages more than one fund at the same time, the
observations are pooled across different funds.
Table 1 provides the summary statistics of the best and worst betas of statistical teams and
actual teams. The median of statistical teams’ betas, β
CF
best (β
CF
worst), is -0.01 (-0.05) and that of actual
teams’ betas, βbest (βworst), is 0.00 (-0.04). The values close to zero suggest that the fund managers
in the sample display diverse cognitive styles when making decisions to sell, some showing a
high degree of rank effect while others show the opposite.
Table 3.6 shows the results of paired t-tests that compare the best and worst betas of statistical
teams and actual teams. Panel (a) shows the results on all teams. Although the betas of actual
93

teams are smaller than those of statistical teams, the differences are not statistically significant at
the 95% confidence level.
To explore if the results arise differently for mutual fund teams with different member compositions in terms of their cognitive styles, I use the marginal effects computed from Equation
(6) to categorize the teams the fund managers join as homogenous and heterogeneous groups. I
define a mutual fund team as homogenous if all team members’ betas have the same sign, either
positive or negative. A homogenous team means that team members show a similar tendency to
use extreme ranks when making decisions to sell in the absence of the team effect. On the other
hand, I define a team as heterogenous if the signs of team members’ individual betas are mixed.
Table 6 Panel (b) shows the results of homogenous teams. Again, the paired t-test results are
not statistically significant, which suggests that teams with members of similar cognitive styles
do not reduce the rank effect. On the other hand, the results of heterogenous teams in Panel (c)
show that both the best and worst betas of actual teams are statistically significantly smaller than
that of statistical teams for cognitively diverse teams.
The results suggest that the cognitive diversity of team members positively affects the collective cognition of teams. Together with the results on common holdings in Section 5.1, the results
suggest that the cue to evoke System 2 is more likely to be generated when fund managers with
different cognitive styles engage in group discussions, during which they are more likely to detect
and correct the errors arising from the use of heuristics.
94

3.6 Conclusion
In this study, I investigate whether the collective cognition of mutual fund teams reduces the rank
effect, a behavioral bias that is suggested to be driven by fund managers’ limited cognitive abilities. I document low post-trade returns of the extreme-ranked stocks that are sold to suggest that
the rank effect is the mistakes caused by fund managers’ indeliberate judgments under System 1
of the dual cognitive theory.
By exploiting data on the full managerial history of the fund managers of actively managed
US equity mutual funds from January 1980 to December 2021, I find that the rank effect arises
smaller in teams than in individuals when the effect of fund management companies and that of
individual fund managers are controlled for. To address the concern with a non-random team
assignment to funds, I employ the propensity score matching method to find consistent results.
By documenting a smaller rank effect for the positions that are more likely to be discussed in
teams and in the teams with cognitive style diversity, I suggest inter-group discussion as the
mechanism that generates the cue to invoke System 2.
Combined, the results suggest that the collective cognition of mutual fund teams displays a
superior cognitive ability than individual fund managers to reduce the rank effect. The results
highlight one advantage of team management over solo management in the mutual fund industry,
the reduction of the use of heuristics when making decisions to sell.
95

Figure 3.1: Team management in the mutual fund industry
This figure plots the percentage of team-managed funds among the actively managed US equity
mutual funds with fund manager information over the sample period from January 1980 to
December 2021.
96

Figure 3.2: Post-trade returns relative to counterfactual
This figure presents the average post-trade returns of extreme-ranked stocks that are sold relative to those of counterfactuals over different time horizons after the quarter trades were made.
Panel (a) uses the post-trade returns of the non-extreme-ranked stocks that are sold as counterfactual returns. Panel (b) uses the returns of a random selling strategy as counterfactual returns,
which is the equal-weighted average of the realized factor-neutral returns across stocks held in
the portfolio. All stock returns are factor-neutral stock returns using Fama-French/Carhart four
factors estimated over the rolling window of the previous sixty months. -1 is multiplied to the
returns of the sold stocks so that positive values indicate that trades help portfolio performance.
(a) RExtremeSell - RN onExtremeSell
(b) RExtremeSell - RHold
97

Figure 3.3: Propensity score matching
This figure plots the results in Table 4 Column (4) with 95% confidence intervals of the difference
between the linear combination of the regression coefficients, separately for the middle-ranked,
best-ranked, and worst-ranked positions.
(a) Non-extreme
(b) Best
(c) Worst
98

Table 3.1: Summary statistics
This table reports the characteristics of the actively managed US equity mutual funds for the sample period from January 1980 to December 2021. β
CF
best and β
CF
worst are the rank effect of statistical
teams, which is the average of team members’ individual rank effect estimated as in Equation
(3.6). βbest and βworst are the measures of rank effect of actual teams.
Statistic Mean St. Dev. Min Pctl(25) Median Pctl(75) Max
Fund monthly TNA ($mil) 933.60 2,259.80 0.20 32.90 145.00 615.30 12,388.04
Team managed dummy 0.67 0.47 0.00 0.00 1.00 1.00 1.00
Num managers 2.84 3.41 0.00 1.00 2.00 3.00 21.00
Manager tenure 10.56 6.81 0.03 5.80 8.85 13.63 47.13
Fund age 12.98 10.53 0.00 6.00 10.00 17.00 64.00
Num holdings 122.94 111.59 20.00 45.00 77.00 161.00 500.00
Monthly fund turnover ratio (%) 0.82 0.67 0.03 0.35 0.67 1.09 4.03
Monthly fund flow 0.001 0.06 −0.19 −0.02 −0.004 0.01 0.47
Fund flow volatility 0.04 0.04 0.002 0.01 0.02 0.05 0.21
Fund monthly return 0.01 0.05 −0.15 −0.02 0.01 0.03 0.14
Carhart alpha −0.001 0.02 −0.15 −0.01 −0.001 0.01 0.21
Carhart alpha volatility 0.01 0.01 0.00 0.01 0.01 0.02 0.12
β
CF
best 0.001 0.14 −0.40 −0.08 −0.01 0.05 0.55
βbest −0.001 0.24 −1.03 −0.07 0.00 0.09 0.63
β
CF
worst −0.13 0.27 −1.94 −0.15 −0.05 −0.01 0.46
βworst −0.20 1.02 −5.28 −0.16 −0.04 0.01 2.08
99

Table 3.2: Rank effect
This table reports the results of the regression that replicates Hartzmark (2015):
Selli,j,t = β0 + β1Besti,j,t + β2W orsti,j,t + StockF Ei × QuarterF Et + ϵi,j,t
where Selli,j,t is a dummy equal to 1 if fund j liquidates (all shares are sold) or sells (a decrease
in the number of shares held) stock i in quarter t, Besti,j,t (W orsti,j,t) is a dummy variable equal
to 1 if stock i has the highest (lowest) rank in the portfolio of fund j in quarter t, when stocks are
ranked based on their return from the purchase price. Returns at the time of sale are calculated
between the purchase price and selling price based on the report dates in the quarterly holding
information. For the stocks that additional shares are purchased after the initiation, the valueweighted average price is used to calculate the purchase price. Stock×Quarter fixed effects to
control for time-varying stock characteristics. Standard errors are clustered by fund and time.
Dependent variable:
Sell dummy
(1) (2) (3)
Best dummy 0.081∗∗∗ 0.097∗∗∗ 0.074∗∗∗
(0.007) (0.007) (0.008)
Worst dummy 0.057∗∗∗ 0.080∗∗∗ 0.041∗∗∗
(0.008) (0.007) (0.010)
Stock × time FE Yes
Stock × time × holding period FE Yes
Stock × time × initiation time FE Yes
Observations 7,514,532 7,514,532 7,514,532
Adjusted R2
0.041 0.179 0.141
100

Table 3.3: Rank effect in teams
This table presents the results of the linear probability model that explores the rank effect in solo
versus team management:
Selli,j,t = β0 + β1Extremei,j,t + β2T eamj,t + β3Extremei,j,t × T eamj,t
+ Numholdingsj,t + α
F E + ϵi,j,t
where Selli,j,t is a dummy equal to 1 if fund j liquidates (all shares are sold) or sells (a decrease
in the number of shares held) stock i in quarter t, Extremei,j,t is a dummy indicating is stock
i is best (worst) ranked in fund j in quarter t, and T eamj,t is a dummy equal to 1 if fund j is
team-managed and 0 if solo-managed. Numholdingsj,t is the number of stock holdings of fund
j in quarter t. Standard errors are clustered by fund and time.
Dependent variable:
Sell dummy
(1) (2) (3) (4) (5) (6)
Best 0.088∗∗∗ 0.065∗∗∗ 0.059∗∗∗ 0.065∗∗∗ 0.059∗∗∗ 0.065∗∗∗
(0.007) (0.007) (0.007) (0.007) (0.007) (0.007)
Worst 0.055∗∗∗ 0.051∗∗∗ 0.040∗∗∗ 0.051∗∗∗ 0.040∗∗∗ 0.051∗∗∗
(0.008) (0.006) (0.007) (0.006) (0.007) (0.006)
Team 0.009∗
0.022∗∗∗ 0.010∗∗∗ 0.022∗∗∗ 0.010∗∗∗ 0.022∗∗∗
(0.005) (0.004) (0.003) (0.004) (0.003) (0.004)
Num holdings −0.000∗∗∗−0.000∗∗∗−0.000∗∗∗−0.000∗∗∗−0.000∗∗∗−0.000∗∗∗
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Best × Team −0.003 −0.024∗∗∗ −0.013∗∗ −0.024∗∗∗ −0.013∗∗ −0.024∗∗∗
(0.006) (0.006) (0.006) (0.006) (0.006) (0.006)
Worst × Team −0.015∗∗ −0.025∗∗∗−0.018∗∗∗−0.025∗∗∗−0.018∗∗∗−0.025∗∗∗
(0.007) (0.006) (0.006) (0.006) (0.006) (0.006)
Stock × time FE Yes Yes
Stock × time × holding period FE Yes Yes
Stock × time × initiation time FE Yes Yes
Fund family FE Yes Yes Yes
Fund manager FE Yes Yes Yes
Observations 2,532,13510,533,3802,532,13510,533,3802,532,13510,533,380
Adjusted R2
0.346 0.647 0.495 0.647 0.495 0.647
101

Table 3.4: Propensity score matching
This table shows the triple difference-in-differences regression results of the rank effect in teams
on the matched samples using the nearest neighbor propensity score matching. For each fund
that switches from solo to team management (treatment group), I identify a control fund with
similar characteristics in the four quarters before the even quarter. To calculate the propensity
score, I use the average quarterly values of TNA, turnover ratio, fund flow, fund flow volatility,
four-factor alpha, and the four-factor alpha volatility. Standard errors are clustered by fund and
time.
Dependent variable:
Sell dummy
(1) (2) (3) (4) (5) (6)
Best 0.051∗∗∗ 0.105∗∗∗ 0.051∗∗∗ 0.105∗∗∗ 0.061∗∗∗ 0.100∗∗∗
(0.013) (0.009) (0.013) (0.009) (0.016) (0.012)
Worst 0.001 0.044∗∗∗ 0.001 0.044∗∗∗ −0.020 0.031∗∗∗
(0.015) (0.009) (0.015) (0.009) (0.015) (0.011)
Treated −0.057∗∗∗−0.056∗∗∗−0.057∗∗∗−0.056∗∗∗−0.053∗∗∗−0.047∗∗∗
(0.018) (0.016) (0.018) (0.016) (0.014) (0.011)
Post −0.023 0.055 −0.023 0.055 −0.028 0.044
(0.041) (0.039) (0.041) (0.039) (0.033) (0.037)
Num holdings −0.001∗∗∗−0.001∗∗∗−0.001∗∗∗−0.001∗∗∗−0.001∗∗∗−0.001∗∗∗
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Best × Treated 0.021 0.018 0.021 0.018 0.022 0.005
(0.016) (0.011) (0.016) (0.011) (0.019) (0.016)
Worst × Treated 0.050∗∗∗ 0.046∗∗∗ 0.050∗∗∗ 0.046∗∗∗ 0.042∗∗∗ 0.028∗∗
(0.016) (0.012) (0.016) (0.012) (0.015) (0.014)
Best × Post 0.076∗∗∗ 0.040∗
0.076∗∗∗ 0.040∗
0.084∗
0.071∗
(0.024) (0.021) (0.024) (0.021) (0.045) (0.038)
Worst × Post 0.129∗∗∗ 0.101∗∗∗ 0.129∗∗∗ 0.101∗∗∗ 0.098∗∗∗ 0.092∗∗
(0.037) (0.033) (0.037) (0.033) (0.037) (0.038)
Treated × Post 0.045 −0.011 0.045 −0.011 0.057 −0.018
(0.044) (0.041) (0.044) (0.041) (0.036) (0.038)
Best × Treated × Post −0.104∗∗∗−0.075∗∗∗−0.104∗∗∗−0.075∗∗∗−0.111∗∗ −0.100∗∗
(0.029) (0.026) (0.029) (0.026) (0.051) (0.042)
Worst × Treated × Post −0.142∗∗∗−0.130∗∗∗−0.142∗∗∗−0.130∗∗∗−0.121∗∗∗−0.097∗∗
(0.041) (0.036) (0.041) (0.036) (0.040) (0.039)
Stock × time FE Yes Yes
Stock × time × holding period FE Yes Yes
Stock × time × initiation time FE Yes Yes
Fund family FE Yes Yes Yes
Observations 2,629,732 2,629,732 2,629,732 2,629,732 2,629,732 2,629,732
Adjusted R2
0.076 0.154 0.076 0.154 0.167 0.236
102

Table 3.5: Rank effect on common holdings
This table presents the results of the regression that compares the rank effect of stocks that are
extreme-ranked in multiple funds that team members manage at the same time and that of the
stocks that are extreme-ranked only in one fund. Only the funds that are team-managed and at
least one of the team members manage other funds at the same time, and the stocks that are also
held by at least one of the team members’ other funds are included in the regression:
Selli,j,t = β0 + β1Besti,j,t + β2CommonBesti,j,t+
β3W orsti,j,t + β4CommonW orsti,j,t + Cj,t + α
F E + ϵi,j,t
Selli,j,t is a dummy equal to 1 if fund j liquidates (all shares are sold) or sells (a decrease in the
number of shares held) stock i in quarter t, Besti,j,t (W orsti,j,t) is a dummy indicating if stock i is
best (worst) ranked in fund j in quarter t. CommonBesti,j,t (CommonW orsti,j,t) is 1 if stock i is
also extreme ranked in other funds that fund j’s fund managers manage in quarter t. For control
variables, Cj,t, the number of holdings of fund j in quarter t and the number of other funds that
team members of fund j manage in quarter t are included. Standard errors are clustered by fund
and time.
Dependent variable:
Sell dummy
(1) (2) (3) (4) (5) (6)
Best 0.079∗∗∗ 0.095∗∗∗ 0.065∗∗∗ 0.066∗∗∗ 0.065∗∗∗ 0.066∗∗∗
(0.006) (0.004) (0.004) (0.004) (0.004) (0.004)
Common Best 0.050∗∗∗ 0.036∗∗∗ 0.006∗
0.003 0.006∗
0.003
(0.005) (0.004) (0.003) (0.003) (0.003) (0.003)
Worst 0.066∗∗∗ 0.085∗∗∗ 0.059∗∗∗ 0.061∗∗∗ 0.059∗∗∗ 0.061∗∗∗
(0.007) (0.005) (0.003) (0.003) (0.003) (0.003)
Common Worst 0.041∗∗∗ 0.019∗∗∗ −0.004 −0.007∗∗ −0.004 −0.007∗∗
(0.006) (0.004) (0.003) (0.003) (0.003) (0.003)
Num connected funds −0.001 −0.000 0.000∗
0.001∗∗∗ 0.0004∗
0.001∗∗∗
(0.001) (0.001) (0.000) (0.000) (0.000) (0.000)
Num holding −0.000∗∗∗−0.000∗∗∗−0.000∗∗∗−0.000∗∗∗−0.000∗∗∗−0.000∗∗∗
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Stock × time FE Yes Yes
Stock × time × holding period FE Yes Yes
Stock × time × initiation time FE Yes Yes
Fund family FE Yes Yes Yes
Observations 5,194,5535,194,5535,194,5535,194,5535,194,5535,194,553
Adjusted R2
0.328 0.414 0.602 0.611 0.602 0.611
103

Table 3.6: Paired t-tests of the rank effect of statistical team versus actual team
This table presents the results of paired t-tests that compare the rank effect of statistical teams
and actual teams. The rank effect of a statistical team is the average of team members’ individual
betas on extreme dummies estimated on their solo-management experiences in the following
specification:
Selli,t = β
j
0 + β
j
BestExtremei,t + β
j
W orstExtremei,t+
β
j
2
(Returni,t × Gaini,t) + β
j
3
(Returni,t × Lossi,t)+
β
j
4Gaini,t + β
j
5
(Returni,t × Gaini,t ×
p
HoldingP eriodi,t)+
β
j
6
(Returni,t × Lossi,t ×
p
HoldingP eriodi,t) + β
j
7
p
HoldingP eriodi,t+
β
j
8
(Gaini,t × V ariancei,t) + β
j
9
(Lossi,t × V ariancei,t) + ϵ
j
i,t
where Selli,t is a dummy variable equal to 1 if fund manager j sells stock i at time t, Besti,t
(W orsti,t) is a dummy variable equal to 1 if stock i is best (worst) ranked in fund j at time t.
Returni,t is a return from purchase and Gaini,t (Lossi,t) is a dummy variable equal to 1 if stock
i’s return from the purchase price is positive (negative) for fund manager j. HoldingP eriod is the
number of months a position is held from the initial purchase quarter. V ariance is the variance of
monthly stock prices during the previous 12 months from quarter t. Homogenous teams in Panel
B are teams in which all team members’ betas have the same sign, either positive or negative.
Heterogenous teams in Panel C are teams where the signs of team members’ individual betas are
mixed.
Panel A. All teams
Statistical Team Mean Differences t-statistic
βBest 0.001 −0.001 −0.001 −0.138
βW orst −0.127 −0.197 −0.068 −1.726
Panel B. Homogenous teams
Statistical Team Mean Differences t-statistic
βBest −0.126 0.115 0.022 1.636
βW orst −0.172 −0.193 −0.018 −0.361
Panel C. Heterogenous teams
Statistical Team Mean Differences t-statistic
βBest 0.023 −0.022 −0.042 −2.465
βW orst −0.037 −0.205 −0.169 −2.720
104

Bibliography
Adams, John C, Takeshi Nishikawa, and Ramesh P Rao (2018). “Mutual fund performance,
management teams, and boards”. Journal of banking & finance 92, pp. 358–368.
Agarwal, Sumit and Robert Hauswald (2010). “Distance and private information in lending”. The
Review of Financial Studies 23.7, pp. 2757–2788.
Aggarwal, Ishani, Anita Williams Woolley, Christopher F Chabris, and Thomas W Malone
(2019). “The impact of cognitive style diversity on implicit learning in teams”. Frontiers in
psychology 10, p. 112.
Ahern, Kenneth R (2017). “Information networks: Evidence from illegal insider trading tips”.
Journal of Financial Economics 125.1, pp. 26–47.
Akepanidtaworn, Klakow, Rick Di Mascio, Alex Imas, and Lawrence Schmidt (2021). Selling fast
and buying slow: Heuristics and trading performance of institutional investors. Tech. rep.
National Bureau of Economic Research.
Antoniou, Constantinos and Shema F Mitali (2018). “Do Stock-Level Experienced Returns Affect
Security Selection?” Available at SSRN 3116073.
Ausburn, Lynna J and Floyd B Ausburn (1978). “Cognitive styles: Some information and
implications for instructional design”. Ectj 26.4, pp. 337–354.
Bae, KeeHong, René M Stulz, and Hongping Tan (2008). “Do local analysts know more? A
cross-country study of the performance of local analysts and foreign analysts”. Journal of
Financial Economics 88.3, pp. 581–606.
Bai, Jennie and Massimo Massa (2022). “Is Hard and Soft Information Substitutable? Evidence
from Lockdown”. Available at SSRN 3843782.
Baik, Bok, Jun-Koo Kang, and Jin-Mo Kim (2010). “Local institutional investors, information
asymmetries, and equity returns”. Journal of financial economics 97.1, pp. 81–106.
105

Bär, Michaela, Alexandra Niessen-Ruenzi, and Stefan Ruenzi (2007). “The impact of work group
diversity on performance: Large sample evidence from the mutual fund industry”. Available
at SSRN 1017803.
Barahona, Ricardo, Stefano Cassella, and Kristy AE Jansen (2022). “Do Teams Alleviate or
Exacerbate Cognitive Biases? Evidence from Extrapolation Bias in Mutual Funds”. Available
at SSRN 3783421.
Bartov, Eli, Lucile Faurel, and Partha S Mohanram (2018). “Can Twitter help predict firm-level
earnings and stock returns?” The Accounting Review 93.3, pp. 25–57.
Ben-David, Itzhak and David Hirshleifer (2012). “Are investors really reluctant to realize their
losses? Trading responses to past returns and the disposition effect”. The Review of Financial
Studies 25.8, pp. 2485–2532.
Ben-Rephael, Azi, Zhi Da, and Ryan D Israelsen (2017). “It depends on where you search:
Institutional investor attention and underreaction to news”. The Review of Financial Studies
30.9, pp. 3009–3047.
Bénabou, Roland (2013). “Groupthink: Collective delusions in organizations and markets”.
Review of economic studies 80.2, pp. 429–462.
Berk, Jonathan B and Jules H Van Binsbergen (2015). “Measuring skill in the mutual fund
industry”. Journal of financial economics 118.1, pp. 1–20.
Bernile, Gennaro, Alok Kumar, and Johan Sulaeman (2015). “Home away from home: Geography
of information and local investors”. The Review of Financial Studies 28.7, pp. 2009–2049.
Bernile, Gennaro, Alok Kumar, Johan Sulaeman, and Qin Wang (2019). “Has Local Information
Advantage Disappeared?” Review of Financial Economics – Special Issue on Behavioral
Finance 37, pp. 38–60.
Bernstein, Shai, Xavier Giroud, and Richard R Townsend (2016). “The impact of venture capital
monitoring”. The Journal of Finance 71.4, pp. 1591–1622.
Bliss, Richard T, Mark E Potter, and Christopher Schwarz (2008). “Performance characteristics
of individually-managed versus team-managed mutual funds”. The Journal of Portfolio
Management 34.3, pp. 110–119.
Brown, Jeffrey R, Zoran Ivković, Paul A Smith, and Scott Weisbenner (2008). “Neighbors matter:
Causal community effects and stock market participation”. The Journal of Finance 63.3,
pp. 1509–1531.
Bushee, Brian J, Michael J Jung, and Gregory S Miller (2011). “Conference presentations and the
disclosure milieu”. Journal of Accounting Research 49.5, pp. 1163–1192.
106

Carhart, Mark M (1997). “On persistence in mutual fund performance”. The Journal of finance
52.1, pp. 57–82.
Chen, Honghui, Yuanyu Qu, Tao Shen, Qinghai Wang, and David X Xu (2022). “The geography
of information acquisition”. Journal of Financial and Quantitative Analysis 57.6,
pp. 2251–2285.
Chen, Joseph, Harrison Hong, Ming Huang, and Jeffrey D Kubik (2004). “Does fund size erode
mutual fund performance? The role of liquidity and organization”. American Economic
Review 94.5, pp. 1276–1302.
Choy, Stacey and Ole-Kristian Hope (2021). “Private Communication between Managers and
Financial Analysts: Evidence from Taxi Ride Patterns in New York City”. Available at SSRN
3920680.
Cici, Gjergji (2012). “The prevalence of the disposition effect in mutual funds’ trades”. Journal of
Financial and Quantitative Analysis 47.4, pp. 795–820.
Cici, Gjergji, Stefan Jaspersen, and Alexander Kempf (2017). “Speed of information diffusion
within fund families”. Review of Asset Pricing Studies 7.1, pp. 144–170.
Cohen, Lauren, Andrea Frazzini, and Christopher Malloy (2008). “The small world of investing:
Board connections and mutual fund returns”. Journal of Political Economy 116.5, pp. 951–979.
— (2010). “Sell-side school ties”. The Journal of Finance 65.4, pp. 1409–1437.
Coval, Joshua and Erik Stafford (2007). “Asset fire sales (and purchases) in equity markets”.
Journal of Financial Economics 86.2, pp. 479–512.
Coval, Joshua D and Tobias J Moskowitz (1999). “Home bias at home: Local equity preference in
domestic portfolios”. The Journal of Finance 54.6, pp. 2045–2073.
— (2001). “The geography of investment: Informed trading and asset prices”. Journal of political
Economy 109.4, pp. 811–841.
Crawford, Steven S, Wesley R Gray, and Andrew E Kern (2017). “Why do fund managers
identify and share profitable ideas?” Journal of Financial and Quantitative Analysis 52.5,
pp. 1903–1926.
Cummings, Jonathon N, Brian Butler, and Robert Kraut (2002). “The quality of online social
relationships”. Communications of the ACM 45.7, pp. 103–108.
Da, Zhi, Umit G Gurun, Bin Li, and Mitch Warachka (2021). “Investment in a smaller world: The
implications of air travel for investors and firms”. Management Science 67.1, pp. 417–435.
107

Daniel, Kent, Mark Grinblatt, Sheridan Titman, and Russ Wermers (1997). “Measuring mutual
fund performance with characteristic-based benchmarks”. The Journal of finance 52.3,
pp. 1035–1058.
Diecidue, Enrico and Peter P Wakker (2001). “On the intuition of rank-dependent utility”.
Journal of Risk and Uncertainty 23.3, pp. 281–298.
Dorn, Daniel and Pramodkumar Yadav (2022). “Team Disposition Effects: Vanity or
Groupthink?” Available at SSRN 4057531.
Dyer, Travis A (2021). “The demand for public information by local and nonlocal investors:
Evidence from investor-level data”. Journal of Accounting and Economics 72.1, p. 101417.
East, Robert, Kathy Hammond, and Malcolm Wright (2007). “The relative incidence of positive
and negative word of mouth: A multi-category study”. International journal of research in
marketing 24.2, pp. 175–184.
Edelen, Roger M, Ozgur S Ince, and Gregory B Kadlec (2016). “Institutional investors and stock
return anomalies”. Journal of Financial Economics 119.3, pp. 472–488.
Ellis, Jesse, Leonardo Madureira, and Shane Underwood (2020). “The causal effects of proximity
on investment: Evidence from flight introductions”. Journal of Financial and Quantitative
Analysis 55.6, pp. 1978–2004.
Engel, David, Anita Williams Woolley, Lisa X Jing, Christopher F Chabris, and
Thomas W Malone (2014). “Reading the mind in the eyes or reading between the lines?
Theory of mind predicts collective intelligence equally well online and face-to-face”. PloS
one 9.12, e115212.
Engelberg, Joseph E and Christopher A Parsons (2011). “The causal impact of media in financial
markets”. The Journal of Finance 66.1, pp. 67–97.
Epstein, Seymour (1994). “Integration of the cognitive and the psychodynamic unconscious.”
American psychologist 49.8, p. 709.
Evans, Richard, Melissa Porras Prado, A Emanuele Rizzo, and Rafael Zambrana (2020). “The
performance of diverse teams: Evidence from US mutual funds”.
Fama, Eugene F and Kenneth R French (1993). “Common risk factors in the returns on stocks
and bonds”. Journal of financial economics 33.1, pp. 3–56.
Fang, Lily H, Joel Peress, and Lu Zheng (2014). “Does media coverage of stocks affect mutual
funds’ trading and performance?” The Review of Financial Studies 27.12, pp. 3441–3466.
Fedyk, Anastassia, Saurin Patel, and Sergei Sarkissian (2023). “Managerial structure and
performance-induced trading”.
108

Frydman, Cary and Ian Krajbich (2022). “Using response times to infer others’ private
information: An application to information cascades”. Management Science 68.4,
pp. 2970–2986.
Giroud, Xavier (2013). “Proximity and investment: Evidence from plant-level data”. The
Quarterly Journal of Economics 128.2, pp. 861–915.
Grosshans, Daniel, Ferdinand Langnickel, and Stefan Zeisberger (2020). “Is buying more
forward-looking than selling? The role of beliefs in investment decisions”. The Role of Beliefs
in Investment Decisions (July 8, 2020).
Han, Bing, David Hirshleifer, and Johan Walden (2018). “Social transmission bias and investor
behavior”. Journal of Financial and Quantitative Analysis, pp. 1–42.
— (2022). “Social transmission bias and investor behavior”. Journal of Financial and
Quantitative Analysis 57.1, pp. 390–412.
Han, Bing and Liyan Yang (2013). “Social networks, information acquisition, and asset prices”.
Management Science 59.6, pp. 1444–1457.
Hartzmark, Samuel M (2015). “The worst, the best, ignoring all the rest: The rank effect and
trading behavior”. The Review of Financial Studies 28.4, pp. 1024–1059.
Harvey, Campbell R, Yan Liu, Eric KM Tan, and Min Zhu (2021). “Crowding: Evidence from fund
managerial structure”. Available at SSRN 3554636.
Hasan, Iftekhar, CHUN-KEUNG HOI, Qiang Wu, and Hao Zhang (2017). “Does social capital
matter in corporate decisions? Evidence from corporate tax avoidance”. Journal of
Accounting Research 55.3, pp. 629–668.
Heimer, Rawley Z (2016). “Peer pressure: Social interaction and the disposition effect”. The
Review of Financial Studies 29.11, pp. 3177–3209.
Hinsz, Verlin B, R Scott Tindale, and David A Vollrath (1997). “The emerging conceptualization
of groups as information processors.” Psychological bulletin 121.1, p. 43.
Hirshleifer, David (2015). “Behavioral finance”. Annual Review of Financial Economics 7,
pp. 133–159.
— (2020). “Presidential address: Social transmission bias in economics and finance”. The
Journal of Finance 75.4, pp. 1779–1831.
Hong, Harrison, Jeffrey D Kubik, and Jeremy C Stein (2004). “Social interaction and
stock-market participation”. The journal of finance 59.1, pp. 137–163.
109

Hong, Harrison, Jeffrey D Kubik, and Jeremy C Stein (2005). “Thy neighbor’s portfolio:
Word-of-mouth effects in the holdings and trades of money managers”. The Journal of
Finance 60.6, pp. 2801–2824.
Hu, Allen and Song Ma (2021). Persuading investors: A video-based study. Tech. rep. National
Bureau of Economic Research.
Huberman, Gur (2001). “Familiarity breeds investment”. The Review of Financial Studies 14.3,
pp. 659–680.
Hvide, Hans K and Per Östberg (2015). “Social interaction at work”. Journal of Financial
Economics 117.3, pp. 628–652.
Janis, I (1982). “L.(1972). Victims of groupthink: A psychological study of foreign-policy
decisions and fiascoes”. Boston (MA).
Jensen, Michael C (1968). “The performance of mutual funds in the period 1945-1964”. The
Journal of finance 23.2, pp. 389–416.
Jiang, Feng, Yiming Qian, and Scott E Yonker (2019). “Hometown biased acquisitions”. Journal of
Financial and Quantitative Analysis 54.5, pp. 2017–2051.
Kacperczyk, Marcin and Amit Seru (2007). “Fund manager use of public information: New
evidence on managerial skills”. The Journal of Finance 62.2, pp. 485–528.
Kahneman, Daniel (1973). Attention and effort. Vol. 1063. Citeseer.
— (2000). “A psychological point of view: Violations of rational rules as a diagnostic of mental
processes”. Behavioral and Brain Sciences 23.5, pp. 681–683.
— (2003). “Maps of bounded rationality: Psychology for behavioral economics”. American
economic review 93.5, pp. 1449–1475.
— (2011). Thinking, fast and slow. Macmillan.
Kahneman, Daniel and Shane Frederick (2004). “Attribute substitution in intuitive judgment”.
Models of a man: Essays in memory of Herbert A. Simon, pp. 411–432.
Kahneman, Daniel, Stewart Paul Slovic, Paul Slovic, and Amos Tversky (1982). Judgment under
uncertainty: Heuristics and biases. Cambridge university press.
Kang, Jung Koo, Lorien Stice-Lawrence, and Yu Ting Forester Wong (2021). “The Firm Next
Door: Using Satellite Images to Study Local Information Advantage”. Journal of Accounting
Research 59.2, pp. 713–750.
110

Kim, Young Ji, DAVID Engel, Anita Williams Woolley, Jeffrey Lin, Naomi McArthur, and
Thomas W Malone (2015). “Work together, play smart: Collective intelligence in League of
Legends teams”. Collective Intelligence Conference.
Kogan, Nathan and Michael A Wallach (1967). “Risky-shift phenomenon in small
decision-making groups: A test of the information-exchange hypothesis”. Journal of
Experimental Social Psychology 3.1, pp. 75–84.
Kozhevnikov, Maria, Stephen Kosslyn, and Jennifer Shephard (2005). “Spatial versus object
visualizers: A new characterization of visual cognitive style”. Memory & cognition 33.4,
pp. 710–726.
Leary, Mark R and Robin M Kowalski (1990). “Impression management: A literature review and
two-component model.” Psychological bulletin 107.1, p. 34.
Liu, Lisa Yao and Shirley Lu (2021). “Information Exposure and Corporate Citizenship”.
Available at SSRN 3757750.
Lu, Yan, Narayan Y Naik, and Melvyn Teo (2024). “Diverse hedge funds”. The Review of Financial
Studies 37.2, pp. 639–683.
Lu, Yan, Sugata Ray, and Melvyn Teo (2016). “Limited attention, marital events and hedge
funds”. Journal of Financial Economics 122.3, pp. 607–624.
Malloy, Christopher J (2005). “The geography of equity analysis”. The Journal of Finance 60.2,
pp. 719–755.
Massa, Massimo, Jonathan Reuter, and Eric Zitzewitz (2010). “When should firms share credit
with employees? Evidence from anonymously managed mutual funds”. Journal of financial
economics 95.3, pp. 400–424.
Mayew, William J and Mohan Venkatachalam (2012). “The power of voice: Managerial affective
states and future firm performance”. The Journal of Finance 67.1, pp. 1–43.
Moscovici, Serge and Marisa Zavalloni (1969). “The group as a polarizer of attitudes.” Journal of
personality and social psychology 12.2, p. 125.
Ng, Lilian, Jing Yu, and Linyang Yu (2021). “COVID-19 Isolation, Managerial Sentiment, and
Corporate Policies”. Managerial Sentiment, and Corporate Policies (April 30, 2021).
Parkinson, Brian and Gwenda Simons (2009). “Affecting others: Social appraisal and emotion
contagion in everyday decision making”. Personality and social psychology bulletin 35.8,
pp. 1071–1084.
Pástor, L’uboš, Robert F Stambaugh, and Lucian A Taylor (2015). “Scale and skill in active
management”. Journal of Financial Economics 116.1, pp. 23–45.
111

Patel, Saurin and Sergei Sarkissian (2017). “To group or not to group? Evidence from mutual
fund databases”. Journal of Financial and Quantitative Analysis 52.5, pp. 1989–2021.
Peng, Lin, Siew Hong Teoh, Yakun Wang, and Jiawen Yan (2021). “Face Value: Trait Impressions,
Performance Characteristics, and Market Outcomes for Financial Analysts”. Available at
SSRN 3741735.
Pool, Veronika K, Noah Stoffman, and Scott E Yonker (2012). “No place like home: Familiarity in
mutual fund manager portfolio choice”. The Review of Financial Studies 25.8, pp. 2563–2599.
— (2015). “The people in your neighborhood: Social interactions and mutual fund portfolios”.
The Journal of Finance 70.6, pp. 2679–2732.
Prather, Larry J and Karen L Middleton (2002). “Are N+ 1 heads better than one?: The case of
mutual fund managers”. Journal of Economic Behavior & Organization 47.1, pp. 103–120.
Puetz, Alexander and Stefan Ruenzi (2011). “Overconfidence among professional investors:
Evidence from mutual fund managers”. Journal of Business Finance & Accounting 38.5-6,
pp. 684–712.
Quinn, Ryan W and Jane E Dutton (2005). “Coordination as energy-in-conversation”. Academy
of management review 30.1, pp. 36–57.
Roberts, John, Paul Sanderson, Richard Barker, and John Hendry (2006). “In the mirror of the
market: The disciplinary effects of company/fund manager meetings”. Accounting,
Organizations and Society 31.3, pp. 277–294.
Schmidt, Daniel (2019). “Distracted institutional investors”. Journal of Financial and Quantitative
Analysis 54.6, pp. 2453–2491.
Seasholes, Mark S and Ning Zhu (2010). “Individual investors and local bias”. The Journal of
Finance 65.5, pp. 1987–2010.
Sialm, Clemens, Zheng Sun, and Lu Zheng (2020). “Home bias and local contagion: Evidence
from funds of hedge funds”. The Review of Financial Studies 33.10, pp. 4771–4810.
Simon, Herbert A (1956). “Rational choice and the structure of the environment.” Psychological
review 63.2, p. 129.
Sloman, Steven A (1996). “The empirical case for two systems of reasoning.” Psychological
bulletin 119.1, p. 3.
Solomon, David and Eugene Soltes (2015). “What are we meeting for? The consequences of
private meetings with investors”. The Journal of Law and Economics 58.2, pp. 325–355.
112

Stanovich, Keith E and Richard F West (2000). “Individual differences in reasoning: Implications
for the rationality debate?” Behavioral and brain sciences 23.5, pp. 645–665.
Stasser, Garold and Zachary Birchmeier (2003). “Group creativity and collective choice”. Group
creativity: Innovation through collaboration, pp. 85–109.
Stoner, James AF (1968). “Risky and cautious shifts in group decisions: The influence of widely
held values”. Journal of Experimental Social Psychology 4.4, pp. 442–459.
Storper, Michael and Anthony J Venables (2004). “Buzz: face-to-face contact and the urban
economy”. Journal of economic geography 4.4, pp. 351–370.
Tosun, Onur Kemal, Liang Jin, Richard Taffler, and Arman Eshraghi (2022). “Fund manager skill:
selling matters more!” Review of Quantitative Finance and Accounting, pp. 1–26.
Tversky, Amos and Daniel Kahneman (1974). “Judgment under Uncertainty: Heuristics and
Biases: Biases in judgments reveal some heuristics of thinking under uncertainty.” science
185.4157, pp. 1124–1131.
Urry, John (2003). “Social networks, travel and talk”. The British journal of sociology 54.2,
pp. 155–175.
Von Cranach, Mario, Guy Ochsenbein, and Ladislav Valach (1986). “The group as a self-active
system: Outline of a theory of group action”. European Journal of social psychology 16.3,
pp. 193–229.
Wegner, Daniel M (1987). “Transactive memory: A contemporary analysis of the group mind”.
Theories of group behavior. Springer, pp. 185–208.
Wittenbaum, Gwen M, Andrea B Hollingshead, and Isabel C Botero (2004). “From cooperative to
motivated information sharing in groups: Moving beyond the hidden profile paradigm”.
Communication Monographs 71.3, pp. 286–310.
Wittenbaum, Gwen M and Garold Stasser (1996). “Management of information in small groups.”
Wojnicki, Andrea C and David Godes (2017). “Signaling success: word of mouth as
self-enhancement”. Customer Needs and Solutions 4.4, pp. 68–82.
Woolley, Anita Williams, Christopher F Chabris, Alex Pentland, Nada Hashmi, and
Thomas W Malone (2010). “Evidence for a collective intelligence factor in the performance
of human groups”. science 330.6004, pp. 686–688.
Yan, Xuemin and Zhe Zhang (2009). “Institutional investors and equity returns: Are short-term
institutions better informed?” The Review of Financial Studies 22.2, pp. 893–924.
113

Yonker, Scott E (2017). “Geography and the market for CEOs”. Management Science 63.3,
pp. 609–630.
Zhu, Qifei (2020). “The missing new funds”. Management Science 66.3, pp. 1193–1204.
114

Abstract (if available)

Abstract This dissertation explores the role of social interactions in shaping fund managers' investment behaviors. Chapter 1 provides a review of the literature. Chapter 2 investigates the causal role of face-to-face communication in generating local informational advantage. By exploiting variation in social interactions driven by COVID-19 lockdowns, I find that during lockdowns, mutual fund managers’ performance on local stocks declined relative to non-local stocks. By establishing causality while ruling out alternative stories based on firm fundamentals and fund managers’ alternative information sources, the results suggest the importance of interpersonal interactions for fund managers to acquire value-relevant information on local stocks. Chapter 3 explores the social element within mutual fund management teams, providing evidence of the superior cognitive ability of mutual fund teams to mitigate a bias driven by constrained cognitive ability. By leveraging data on the managerial history of mutual fund managers, I find a smaller rank effect in team-managed funds relative to solo-managed funds. This reduction is particularly pronounced for stocks that team members are more likely to discuss and in teams with cognitive style diversity. The results suggest a beneficial role of team decision-making in reducing reliance on heuristics when making investment decisions.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Essays on the firm and stakeholders relationships: evidence from mergers & acquisitions and labor negotiations

PDF

Two essays on the mutual fund industry and an application of the optimal risk allocation model in the real estate market

PDF

Understanding the disclosure practices of firms affected by a natural disaster: the case of hurricanes

PDF

Essays in empirical asset pricing

PDF

Essays on real options

PDF

Evolution of returns to scale and investor flows during the life cycle of active asset management

PDF

Essays on the effect of cognitive constraints on financial decision-making

PDF

Two essays on financial econometrics

PDF

Mutual fund screening versus weighting

PDF

Shareholder litigation as a disciplining device: evidence from firms' financial policies

PDF

Essays on delegated portfolio management under market imperfections

PDF

Three essays on behavioral finance with rational foundations

PDF

Essays in financial intermediation

PDF

Essays in asset pricing

PDF

Insider editing on Wikipedia

PDF

Essays in corporate finance

PDF

Essays on information and financial economics

PDF

Management practices of social wealth funds: an exploratory study

PDF

Social movements and access to credit

PDF

The morality of technology

Asset Metadata

Creator Lee, Robin (author)

Core Title Investment behavior of mutual fund managers

School Marshall School of Business

Degree Doctor of Philosophy

Degree Program Finance

Degree Conferral Date 2024-05

Publication Date 05/29/2024

Defense Date 04/23/2024

Publisher Los Angeles, California (original), University of Southern California (original), University of Southern California. Libraries (digital)

Tag behavioral finance,mutual fund,OAI-PMH Harvest,social finance

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Ahern, Kenneth (committee chair), Dechow, Patricia (committee member), Hirshleifer, David (committee member), Hoberg, Gerard (committee member)

Creator Email robin.lee@marshall.usc.edu,robinyou@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC113984882

Unique identifier UC113984882

Identifier etd-LeeRobin-13045.pdf (filename)

Legacy Identifier etd-LeeRobin-13045

Document Type Dissertation

Format theses (aat)

Rights Lee, Robin

Internet Media Type application/pdf

Type texts

Source 20240603-usctheses-batch-1164 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu