ESSAYS IN FINANCIAL ECONOMICS

By

Alexander Ferko

A DISSERTATION

Michigan State University

in partial fulﬁllment of the requirements

Submitted to

for the degree of

Business Administration - Finance – Doctor of Philosophy

2019

ABSTRACT

ESSAYS IN FINANCIAL ECONOMICS

By

Alexander Ferko

Stress negatively aﬀects decision-making processes and limits individuals’ ability to concentrate
or learn new information. Therefore, investors aﬀected by stress experience an increase in their
marginal costs of processing economic information and, consequently, face greater diﬃculty in
making investment allocation decisions. Medical and epidemiology literature document increased
levels of Post-Traumatic Stress Disorder within the general population in the aftermath of widely
reported disaster events. Accordingly, I construct a time-series sample of nationally signiﬁcant
traumatic events such as school shootings to identify periods of elevated stress in the United States
in the period from 1995 to 2017 and study the eﬀects of stress on investors’ ﬁnancial decision-
making abilities.

In this dissertation, I exploit geographic variation in the level of stress exposure across mutual-
fund managers and ﬁnd that managers in close proximity to traumatic events underperform their
geographically distant peers over the quarter following a nearby traumatic event by approximately 50
basis points. I also ﬁnd that, following traumatic events, investors experience increased sensitivity to
search costs, lower sensitivity to ﬁnancial information, and decreased ability to pick mutual funds
that generate positive risk-adjusted returns. These ﬁndings are especially pronounced for retail
investors, who likely face greater information-processing constraints than ﬁnancially sophisticated
institutional investors do.

Copyright by
ALEXANDER FERKO
2019

This thesis is dedicated to my wife, Qingqiu Li.

iv

ACKNOWLEDGEMENTS

I would like to thank my dissertation advisor, Zoran Ivković, for the many helpful comments. I
also received numerous very helpful comments from Andrei Simonov, Charles Hadlock, and Xing
Huang. I would also like to thank the participants of the workshop at Michigan State University.
All remaining errors are my own.

v

TABLE OF CONTENTS

.
.

.
.

.
.

.
.

.
.
.
.

.
.

.
.
.
.

.
.

.
.
.
.

.
.

.
.
.
.

.
.

.
.
.
.

CHAPTER 1

2.2.1
2.2.2 Additional Results

LIST OF TABLES .
LIST OF FIGURES .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
INTRODUCTION .
1
2
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.
1.1 Fund Managers
4
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2 Fund Managers
6
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3 Combined .
.
7
1.4 Background .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.5 Related Work: Investor Psychology and Attention . . . . . . . . . . . . . . . . . .
9
1.6 Sample Construction and Methodology . . . . . . . . . . . . . . . . . . . . . . . . 11
1.6.1 Mutual Fund Variables and Data . . . . . . . . . . . . . . . . . . . . . . . 12
1.6.2 Key Variables: Choice of Events . . . . . . . . . . . . . . . . . . . . . . . 13
CHAPTER 2 FUND MANAGERS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.1 Motivation and Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2 Fund Manager Results .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Primary Results: Fund Returns . . . . . . . . . . . . . . . . . . . . . . . . 19
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2.2.1 Risk Exposure . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Fund Turnover
2.2.2.2
. . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2.2.3
Fund Indexing Strategies . . . . . . . . . . . . . . . . . . . . . . 25
2.3 Alternative Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
CHAPTER 3 MUTUAL FUND CLIENTS . . . . . . . . . . . . . . . . . . . . . . . . . . 28
. 28
3.1 Mutual Fund Client Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
Fund Visibility .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
Flow-Performance Sensitivity . . . . . . . . . . . . . . . . . . . . . . . . 31
. . . . . . . . . . . . . . . . . . . . 32
. 34
Share Purchases and Redemptions . . . . . . . . . . . . . . . . . . . . . . 36
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
.
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
FUND MANAGERS . . . . . . . . . . . . . . . . . . . . . . . . . 44
MUTUAL FUND CLIENTS . . . . . . . . . . . . . . . . . . . . . 60
.
. 75

3.1.1
3.1.2
3.1.3 Retail and Institutional Share-Classes
3.1.4 Decision Making Ability . . . . . . . . . . . . . . . . . . . . . . . . . .
3.1.5

3.2 Robustness Checks
.

.

APPENDICES .
APPENDIX A
APPENDIX B
APPENDIX C
BIBLIOGRAPHY .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3.0.1 Methodology .

.

.

.

.

.

.

.

.

.

.
.

.
.

.
.

.

.

.

.

vi

LIST OF TABLES

Table 2.1: Summary Statistics .

.

.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

Table 2.2: Summary Statistics: Matched Sample . . . . . . . . . . . . . . . . . . . . . . . 51

Table 2.3: Relative Daily Performance of Local and Distant Mutual Funds Following

Stress Events . .

. .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

Table 2.4: Relative Abnormal Performance of Local and Distant Mutual Funds Following

Stress Events . .

. .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

Table 2.5: Mutual Funds Change in Risk Exposure Following Traumatic Events

. . . . . . 55

Table 2.6: Mutual Fund Change in Turnover Following Traumatic Events . . . . . . . . . . 57

Table 2.7: Mutual Fund Tracking Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

Table 3.1: Summary Stats .

.

.

.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

Table 3.2: Mutual Fund Flow Sensitivity to Search Costs During Normal and Stress Periods

63

Table 3.3: Mutual Fund Flow Sensitivity to Past Performance During Normal and Stress

Periods

. .

. .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

.

.

.

.

.

.

Table 3.4: Retail Share-Class Sensitivity to Past Performance During Normal and Stress

Periods

. .

. .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

Table 3.5: Smart Money: 24 month Fund Flow Weighted Alphas after Traumatic Events . . 69

Table 3.6: NSAR: Stress Eﬀects on Mutual Fund Inﬂow and Outﬂows . . . . . . . . . . . . 70

Table 3.7: Relative Daily Performance of Local and Distant Mutual Funds Following

Stress Events . .

. .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

Table 3.8: Robustness Check: Out of Sample Events . . . . . . . . . . . . . . . . . . . . . 73

Table 3.9: Fund Flows - Early vs late period . . . . . . . . . . . . . . . . . . . . . . . . . . 74

vii

LIST OF FIGURES

Figure 1.1: The frequency distribution of events by year

. . . . . . . . . . . . . . . . . . . 41

Figure 1.2: Event Month Formation Rules . . . . . . . . . . . . . . . . . . . . . . . . . .

. 42

Figure 1.3: US Event Locations .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

Figure 2.1: Cumulative Event Time Returns for International Equity Funds . . . . . . . . . 45

Figure 2.2: Cumulative Near vs Distant Simulated Mutual Fund Performance . . . . . . . . 46

Figure 2.3: Cumulative Near vs Distant Simulated Abnormal Mutual Fund Performance .

. 48

Figure 3.1: Flow-Performance Sensitivity: Normal vs Stress . . . . . . . . . . . . . . . .

. 61

viii

CHAPTER 1

INTRODUCTION

The study of behavioral ﬁnance has led to many insights into the biases that govern how individuals
make ﬁnancial decisions.Thaler and Sunstein (2008) suggest that these biases regularly lead indi-
viduals to make sub-optimal economic decisions with incomplete information. Recent literature has
examined how investment decisions vary with investor psychology and emotion ( Goetzmann et al.
(2015) and Lo et al. (2005)). Stress is a widely experienced psychological mechanism that nega-
tively impairs decision-making processes. According to the American Psychological Association’s
2015 Stress in America Survey, 24% of adults experienced ‘extreme’ levels of stress during the past
year, 27% of adults reported an inability to concentrate due to stress in the past month, and 1/3
of respondents reported feelings of anxiety, worry, and a lack of patience(American Psychological
Association, 2016). Investor stress has implications for ﬁnancial markets because investors experi-
encing stress will have greater diﬃculty in making investment allocation decisions. Furthermore,
environmental shocks that aﬀect individual levels of stress may have economic consequences.

In this study, I explore how post-traumatic stress disorder aﬀects investor decision-making within
the mutual-fund sector. Medical and psychological research has shown that post-traumatic stress
directly impairs the cognitive abilities related to decision-making by reducing concentration and
analytical abilities (Vasterling, Brailey, Constans, and Sutker (1998) and Vasterling, Duke, Brailey,
Constans, Allain, and Sutker (2002)). Slone (2000) and Ahern, Galea, Resnick, and Kilpatrick
(2002) ﬁnd evidence of high levels of PTSD symptoms within the general public following traumatic
events, with individuals who consume television coverage of these events being particularly aﬀected.
Based on this evidence, I conjecture that stress and post-traumatic stress symptoms increase investor
information-processing costs by reducing the investors’ ability to concentrate and acquire new
ﬁnancial information. Financial decision-makers suﬀering from post-traumatic stress face increased
diﬃculty in analyzing and utilizing ﬁnancial information, resulting in impaired investment decision
abilities. Subsequently, investment decisions made during high-stress periods are likely to be less

1

informed.

I construct a data set of national traumatic events, such as school shootings, to identify periods of
elevated post-traumatic stress symptoms within the general public. I ﬁnd that, in periods associated
with high PTSD, investors suﬀer from impaired decision-making ability. Mutual-fund managers
located near the sites of traumatic events underperform compared to their distant peers.
I also
ﬁnd that investors have increased sensitivity to search costs and are less sensitive to ﬁnancial
information. These ﬁndings are especially pronounced for retail investors, who likely face greater
information processing constraints than more ﬁnancially sophisticated institutional investors do.
Furthermore, I ﬁnd evidence of a reduction in investors’ mutual-fund selection, with the fund’s
investors choose earning lower alphas.

The traumatic events I use in this study are based on the Vanderbilt Television News Archive.
This archive contains a listing of nightly evening news broadcasts from major national networks
such as CNN, Fox, ABC, NBC, and CBS. The events in my set are highly tragic, sensational,
and dominate the evening news coverage when they occur. They include school shootings, plane
crashes, celebrity deaths, and bombings. Importantly, these events occur randomly and are largely
independent of aggregate market information.

1.1 Fund Managers

In my ﬁrst study, I investigate how post-traumatic stress aﬀects the decision-making of ﬁnan-
cially sophisticated mutual-fund managers. I exploit geographic variation in the exposure of fund
managers to the stress shocks to identify the eﬀects of PTSD on ﬁnancial decision-making abilities.
Psychology research has documented that geographic proximity to traumatic events is related to
an increased exposure to acute post-traumatic stress. (Gabriel et al. (2007), Rubin et al. (2005)).
When a traumatic event exogenously shocks a geographic region, local mutual-fund managers will
experience a higher incidence of post-traumatic stress than distant mutual-fund managers do.

I ﬁnd evidence of reduced portfolio performance within the subsample of managers local to the
traumatic event. This underperformance amounts to 56 basis points over the 3 months following

2

the event for domestic equity mutual-fund managers when benchmarked against a matched sample
of funds located more than 1,000 miles away from the event. To rule out the possibility of local
information eﬀects (Shive (2011)) or any potential home bias (Pool et al. (2012)), I restrict this
sample to include only international equity funds located in the United States and ﬁnd a similar
40-basis point underperformance over the following quarter. In simulations of randomized event
days, I ﬁnd that my results are robust at the 5% level.

Managers may respond to this shock to their marginal information processing costs by ei-
ther delaying investment decisions (lowering decision frequency) or by making relatively ‘easier’
decisions (changing investment strategy). To test this ﬁrst channel, I proxy investment decision
frequency with mutual-fund turnover. If psychological stress is a limiting factor on managers ability
to make decisions, managers may reduce their decision volume and shift new investments into a
future period, resulting a lower fund turnover. I ﬁnd that, following traumatic events, local fund
managers reduce fund turnover relative to distant managers. On an annual basis, this reduction in
turnover amounts to 20% of fund assets for the local managers, while distant managers experience
no observable decrease in annual turnover.

Next, I explore whether the managers’ investment strategy changes in the aftermath of traumatic
events. When making investment decisions, managers strike a trade-oﬀ between the extent to which
they will replicate their benchmark index and the extent to which they will actively deviate from the
index (Cremers and Petajisto (2009)). Replicating the benchmark index is a relatively easier task
because the portfolio positions and weights are known in advance. On the other hand, deviating from
the index requires considerable eﬀort to identify promising investment opportunities. I calculate
tracking errors between each fund and their respective benchmark and ﬁnd that managers who were
more inclined to track their benchmark prior to traumatic events further increase their indexing
behavior and have lower tracking errors in the post-event period.

The ﬁnance literature has proposed two alternative hypothesis for a shift in manager behavior
following traumatic events. The ﬁrst is that the events may change fund managers’ risk aversion.
This risk-aversion hypothesis has been at the core of research into the market reactions to terrorist

3

activities in studies such as Antoniou et al. (2018), Wang, Albert and Young, Michael (2018),
and Wang and Young (2018). To rule out that post-event performance is aﬀected by shifts in risk
aversion, I measure the changes to Fama-French 3-Factor exposures between the pre-event and
post-event quarters. I ﬁnd no evidence of a shift in mutual funds’ risk-factor exposures, suggesting
that changes in risk aversion do not play a role in this context. The second alternative hypothesis
I consider is that managers may be distracted by the traumatic events. Managers have limited
attention capabilities and may reallocate their attention from ﬁnancial information to following
the traumatic events. Similarly, in an eﬀort to boost their coverage of the traumatic events and
thereby increase viewership, news programs may disregard market-relevant information events. If
the results were driven by the distraction while the event was in the news-cycle, the bulk of the
measured eﬀects would occur contemporaneously. However, I ﬁnd that PTSD impairment aﬀects
trading behavior up to 60 trading days following the initial event.

1.2 Fund Managers

In a parallel set of results, I study mutual-fund ﬂows—the observable outcomes of investor
wealth-allocation decisions, to investigate how these events aﬀect mutual-fund investors’ ﬁnancial
decisions. I test for cross-sectional diﬀerences in of fund ﬂows between normal periods and high-
stress periods. Changes in how investors allocate their wealth across diﬀerent types of mutual funds
indicate a shift in how ﬁnancial decisions are made in the aftermath of traumatic events. In other
tests, I examine how stress aﬀects more ﬁnancially sophisticated market participants, the mutual-
fund managers. The availability of daily mutual-fund returns allows for repeated observations of
individual managers’ performance over time.

I ﬁrst examine whether stress impairs the investors’ ability to identify investment opportunities.
In a high-stress environment, in which investors’ ability to concentrate is impaired, managing
mutual-fund search costs likely plays a more pronounced role in determining fund allocation
decisions. Low-visibility funds require more costly eﬀort to identify and are less likely to enter the
choice set of individual investors than large more prominent funds are. I ﬁnd that high-visibility

4

funds receive a greater portion of fund ﬂows during high stress periods. The expected net ﬂows to
the largest funds are approximately 1% greater after traumatic periods.

I next study whether post-traumatic stress limits the amount of information investors incorporate
into their mutual-fund investment decisions. A large body of literature has found that investors
chase past returns in mutual funds by investing disproportionately in the funds that outperformed
in the prior period (Chevalier and Ellison (1997), Sirri and Tufano (1998), Lynch and Musto
(2003),Jonathan B. Berk and Richard C. Green (2004)). To implement the strategy of chasing
past winners, individuals must ﬁrst exert eﬀort to gather, analyze, and compare ﬁnancial infor-
mation. The eﬀort required to learn and process such information is greater in the aftermath of
traumatic events because of increased levels of stress. I estimate the sensitivity of mutual-fund
ﬂows to past returns and ﬁnd that the ﬂow-performance sensitivity is reduced in periods following
traumatic events. The expected ﬂows to the highest-performing funds is almost halved, with the
top-performing funds receiving only 2.5% higher ﬂows compared to 4% for the normal period.
This evidence suggests that investors weigh past return information less heavily during high stress
periods.

Prior familiarity with ﬁnancial information may mitigate the eﬀects of stress on ﬁnancial
decision-making. Unlike institutional investors, retail investors often face barriers in accessing
mutual-fund return information and may lack prior experience in analyzing mutual funds. These
investors need to compare prospectus statements or access premium subscriptions to online tools
such as Morningstar. On the other hand, sophisticated institutional investors face virtually no
barriers in accessing this type of ﬁnancial information. Stress may have a greater eﬀect on retail
investors who face larger marginal increases in information processing costs following traumatic
events. To test this hypothesis, I split my sample of mutual funds and separately estimate ﬂow-
performance sensitivities within retail and institutional fund share classes. I only ﬁnd evidence of a
ﬂatter ﬂow-performance sensitivity within the retail share-class group, conﬁrming my hypothesis.
My next hypothesis addresses whether the quality of investment decisions is reduced by stress.
If investors eﬃciently use ﬁnancial information to allocate their wealth, an increase in information

5

costs during high stress periods will lead to worse investment decisions and reduced portfolio
alphas. In the spirit of Zheng (1999), I construct a monthly portfolio representing the aggregate
decision-making ability of investors. This portfolio weighs the returns of mutual funds that
investors purchased (large inﬂows) less the returns of funds that experienced share redemptions
(large outﬂows). I test the diﬀerence in portfolio alphas between funds purchased and funds sold
for normal and stress periods and perform a series of simulations using randomized event days.
For retail investors, I ﬁnd no evidence of decision-making impairment during high stress periods.
However, for institutional investors, I ﬁnd a decrease in expected alpha of 17 basis points. One
interpretation of this evidence is that institutional investors eﬃciently use ﬁnancial information and
an increase in stress limits their ability to optimally allocate portfolio capital.

Changes in the pattern of net fund ﬂows could be driven by distortions in either purchases of
fund shares (inﬂows) or share redemptions (outﬂows). Investors may respond to the shock of a
traumatic event by delaying their investment decisions. However, investors may have limited ability
to delay share redemptions because these decisions are more likely to be motivated by liquidity
reasons (Barber and Odean, 2000) making share purchases more sensitive to the cognitive eﬀects
of stress. I test this hypothesis using EDGAR Form N-SAR ﬁlings, which separately report fund
inﬂows and outﬂows. I only ﬁnd evidence of a reduction of share purchases following traumatic
events and no eﬀect for share redemptions.

1.3 Combined

My paper explores the role investor psychology has in ﬁnancial decision-making and investment
allocations. I identify environmental shocks to stress as a separate dynamic channel for investor
psychology to aﬀect ﬁnancial markets. My work is related to the study of how environmental
factors aﬀect investor behavior. Kamstra, Kramer, and Levi (2003) and Kamstra, Kramer, Levi,
and Wermers (2017) analyze the eﬀects of Seasonal Aﬀective Disorder on market prices and fund
ﬂows. Investor emotions have been widely studied within the ﬁnance literature in studies such as
Hirshleifer and Shumway (2003), Edmans, García, and Norli (2007), and Goetzmann, Kim, Kumar,

6

and Wang (2015).

The connection between traumatic events and ﬁnancial markets is a growing area of ﬁnancial
research. Antoniou, Maligkris, and Kumar (2018) studies the forecasts of ﬁnancial analysts in the
quarter after domestic terrorist attacks and ﬁnds evidence of increased forecast pessimism. Ahern
(2018) studies terrorist attacks at the institutional and macroeconomic level. Wang and Young
(2018) uses household-level ﬁnancial data and documents reduced participation in equity markets
following terrorist attacks. My dissertation builds upon this literature by connecting the changes
in investor behavior to post-traumatic stress related symptoms and documenting how stress eﬀects
diﬀerent types of investor groups.

Fund ﬂows have been widely studied within the literature in papers such as Elton, Gruber,
and Busse (2004), Cooper, Gulen, and Rau (2005), Cronqvist (2006), Sensoy (2009), and Barber,
Huang, and Odean (2016). My dissertation is also related to the cross-section of mutual-fund ﬂows.
I contribute to this literature by showing how fund allocations are sensitive to changes in investor
psychology.

1.4 Background

Post-traumatic Stress Disorder is the most common type of mental disorder experienced after
large traumatic events Neria, DiGrande, and Adams (2011). PTSD is deﬁned by the DSM-IV
(American Psychiatric Association, 2000) criteria as mental disorder which develops after exposure
to a catastrophic event involving death or injury or perceived threat to one’s physical integrity.
Importantly, individuals can develop symptoms of PTSD without physically being present at a
traumatic event. PTSD only necessitates that individuals perceive a threat to their well-being.
Individuals suﬀering PTSD symptoms commonly experience: substantial functional impairment;
depression and generalized anxiety disorder; feelings of helplessness, avoidance or numbing; and
hyper-arousal (lack of focus). These symptoms can persist for more than four weeks. Silver,
Holman, McIntosh, Poulin, and Gil-Rivas (2002) ﬁnds that individuals often cope with PTSD
symptoms by ‘giving up.’

7

Post-Traumatic Stress Disorder symptoms are directly related to individual decision-making
processes. Vasterling et al. (1998) and Vasterling et al. (2002) study the learning abilities of
Vietnam veterans suﬀering from PTSD and ﬁnd that PTSD reduces learning abilities by impairing
sustained attention, working memory, and initial acquisition of information. Shaw, Applegate, and
Schorr (1996) examine the performance of children in classrooms 21 months following Hurricane
Andrew and ﬁnd evidence of increased attention problems and signs of moderate to severe stress-
related symptoms. Furthermore, PTSD symptoms are psychologically related to those of ADHD
because both disorders share similar symptoms of hyper-arousal (lack of focus). A few papers in
this area of literature, including Szymanski, Sapanski, and Conway (2011), have studied whether
childhood ADHD may be misdiagnosed PTSD.

A growing body of literature has emerged studying the incidence of PTSD symptoms within
the general population following widely reported traumatic events. There is strong evidence
of high levels of PTSD symptoms within the general U.S. population outside of New York City
following the September 11th attacks (Galea, Vlahov, Resnick, Ahern, Susser, Gold, Bucuvalas, and
Kilpatrick (2003), Silver, Holman, McIntosh, Poulin, and Gil-Rivas (2002), Silver et al. (2004),
and Schuster et al. (2001)). These studies ﬁnd that approximately 44% of the U.S. population
reported substantial stress reactions in the ﬁrst week following the attacks. 17% of the population
outside of New York City reported post-traumatic stress symptoms two months after the attacks.
Television news coverage is an important channel for the transmission of these psychological eﬀects.
Notably, Schuster et al. (2001), Ahern et al. (2002), Schlenger, Caddell, Ebert, Jordan, Rourke,
Wilson, Thalji, Dennis, Fairbank, and Kulka (2002), and Slone (2000) ﬁnd that post-traumatic
stress symptoms following the September 11th attacks are increasing with the number of hours of
TV news viewed. In an economic setting, Melnick and Eldor (2010) ﬁnds larger economic eﬀects
to terrorism following increased media coverage of the event.

The emergence of PTSD symptoms within the general population has been studied for traumatic
events other than the September 11th terrorist attacks. Terr, Bloch, Michel, Shi, and al (1999) studies
the eﬀects of PTSD in school age children following the 1989 Challenger Space Shuttle explosion

8

and ﬁnds that East Coast children who watched the explosion exhibited PTSD symptoms in greater
frequency than West Coast children who heard about the event afterwards. The authors also ﬁnd
no diﬀerence in symptoms after 5-7 weeks between students who watched the event on television
and students who were present at Cape Canaveral. Similar patterns emerge for the Oklahoma
City bombing (Pfeﬀerbaum, Doughty, Reddy, Patel, Gurwitch, Nixon, and Tivis, 2002), the 2005
London Subway Bombings (Gabriel, Ferrando, Cortón, Mingote, García-Camba, Liria, and Galea,
2007), and the 2004 Madrid Train Bombings (Rubin et al., 2005).

This line of medical and psychological research suggests a long-lived lingering psychological
impairment following the televised reporting of traumatic events. Based upon these past ﬁndings,
I conjecture that individual information-processing costs are increased during periods of elevated
post-traumatic stress. A reduced ability to concentrate and acquire new information prompts
individuals to have limited ability to process ﬁnancial information before feeling overwhelmed.
The aﬀected individuals must exert more eﬀort to process the same amount of information. As a
result, investment decisions made during high-stress periods are likely to reﬂect less information,
leading to distortions of the cross-section of fund ﬂows and reduced investment performance.
These mutual-fund allocations may be long-lived because individuals infrequently revisit retirement
portfolio allocation decisions (Sialm, Starks, and Zhang, 2015).

1.5 Related Work: Investor Psychology and Attention

The connection between psychological symptoms and ﬁnancial decision-making has been well
established in the ﬁnancial literature. The underlying psychological mechanisms of emotions and
mood have been studied more generally in Lo and Repin (2002) and Kuhnen and Knutson (2011),
who ﬁnd that physiological emotional responses aﬀect ﬁnancial decisions. A large literature has
emerged studying how ﬁnancial markets are aﬀected by environment-induced changes in investor
psychology and moods. Kamstra, Kramer, and Levi (2003) and Kamstra, Kramer, Levi, and
Wermers (2017) analyze the eﬀects of Seasonal Aﬀective Disorder on market prices and fund ﬂows.
The authors use changes in seasons to identify aggregate levels of depression-related symptoms

9

within the investor population and measure subsequent eﬀects on market prices and decisions.
Hirshleifer and Shumway (2003) and Goetzmann, Kim, Kumar, and Wang (2015) explore daily
variation in weather and the resulting eﬀects of changes in investor mood. Edmans, García, and
Norli (2007) ﬁnds similar results using the changes in investor mood following national sporting
events. I build upon this literature by providing evidence that exogenous environmental shocks
in the national media can spill over into ﬁnancial markets. This is a potentially cleaner setting to
study environmental shocks to investor psychology because weather and seasonal trends are fairly
predictable within geographic regions.

My paper is also related to the growing literature studying the economic eﬀects of terrorist
activities. Similar to prior literature on investor moods, Drakos (2010) and Arin, Ciferri, and
Spagnolo (2008) document negative returns attributable to a negative sentiment and mood in
investors following terrorist activity. Melnick and Eldor (2010) ﬁnd that these economic eﬀects are
greater when there is more media coverage of the event. Antoniou, Maligkris, and Kumar (2018)
report that ﬁnancial analysts are more pessimistic in their earnings forecasts following terrorist
activity. Terrorism may also lead to an increased risk aversion among household investors. Building
on Guiso, Sapienza, and Zingales (2018), Wang and Young (2018) ﬁnds reduced participation in
equity markets following terrorist activities and Wang, Albert and Young, Michael (2018) report
evidence that investors shift investment preferences towards less risky mutual funds. Ahern (2018)
studies how psychological stress is connected to economic activity; the paper explores the link
between psychology and the macro-level economy and ﬁnds an overall increase in income following
terrorist activity. This dissertation identiﬁes a separate channel for terrorist activities and traumatic
events to aﬀect investor psychology through PTSD and stress. Whereas other studies focus on
investor risk aversion following terrorist attacks, this PTSD channel aﬀects ﬁnancial markets
through an increase in marginal information costs, reducing the market participants’ investment
ability.

My dissertation is also related to investor attention, a topic widely studied in the ﬁnancial
literature. Peng (2005), DellaVigna and Pollet (2009), Hirshleifer, Lim, and Teoh (2009), Kempf,

10

Manconi, and Spalt (2017) look at the immediate eﬀects of investor attention at the daily level.
Lu, Ray, and Teo (2016) study the performance of hedge-fund managers during marital events. In
these papers, there are constraints on investors’ information-processing capacity. My dissertation
is related to this strand of research because stress directly aﬀects investor’s concentration and
information consumption. Israeli, Kasznik, and Sridharan (2017) and Peress and Schmidt (2016)
use news media reporting as an attention grabbing event and study the subsequent eﬀects on trading
volume and liquidity. Rather than study the contemporaneous market eﬀects of attention during
the reporting period, this dissertation examines how stress generated by news media events can
lead to longer-term eﬀects. PTSD can aﬀect individuals over a period of one to two months
(Silver, Holman, McIntosh, Poulin, and Gil-Rivas, 2002), which can lead to stress symptoms and
information inattention long after the initial event seizes to be reported in the news media.

I also contribute to the literature exploring investor ﬁnancial sophistication and diﬀerences in
investment decisions between retail and institutional investors, building upon the literature studying
individual investors (Barber and Odean (2000) and Barber and Odean (2008)). Recent papers
by Foerster, Linnainmaa, Melzer, and Previtero (2017) and Linnainmaa, Melzer, and Previtero
(2017) examine how sophisticated ﬁnancial advice aﬀects investment decisions by unsophisticated
retail investors. Other work by Barber, Huang, and Odean (2016) ﬁnds that sophisticated and
unsophisticated investors exhibit diﬀerent preferences for investment risk when choosing between
mutual funds.

1.6 Sample Construction and Methodology

The data for my analysis comes from two primary sources. Data on mutual funds characteristics,
fund ﬂows, and returns is from the Center for Research in Security Prices (CRSP). Data on media
events comes from the Vanderbilt Television News Archive, an online database containing program
descriptions for nightly news broadcasts by nationally broadcast news stations since 1968. This
data set contains program descriptions for the ABC, CBS, NBC, CNN, MSNBC, and Fox News
networks. Additional data on mutual fund share purchases and redemptions is from SEC From

11

N-SAR, available from EDGAR.

1.6.1 Mutual Fund Variables and Data

Using the CRSP mutual fund database, I construct a sample of mutual funds for the period from
1995 to 2016.
I ﬁrst use a mapping ﬁle provided by CRSP to link CRSP fund identiﬁers into
EDGAR contract, series, and company CIK identiﬁers. This allows me to aggregate the CRSP
share classes into EDGAR fund series (a single fund may have separate share classes for retail,
institutional, and broker sales). My ﬁrst set of tests is conducted at the aggregated fund-series
level. I proportionally weigh CRSP fund characteristics based on monthly total net assets of each
share class within the mutual fund. I apply a standard set of ﬁlters to construct my ﬁnal sample
used in my tests. I require funds to be at least 2 years old, have at least 10 million in monthly total
net assets, have monthly ﬂows of less than 10,000% and ﬂows greater than -97%, and at least 24
monthly-return observations in my data. I further restrict the sample to remove both VAU funds
and ETFs. To reduce the eﬀect of outliers, monthly return data is winsorized at the 1% level.
Furthermore, the fund share classes in CRSP must have existing mappings into the EDGAR data to
identify the fund series and the fund family to which they belong. My fund manager tests use this
sample with the inclusion of the daily mutual fund returns data from CRSP. However, daily mutual
fund returns are only available post-2001. Furthermore, I restrict the mutual-fund manager sample
to only include domestic or international equity funds. This step ensures that I am able to compare
funds along measures of risk exposures and risk adjusted performance. To construct this sample, I
require that the ﬁrst letter of the CRSP objective identiﬁer begins with E.

Panel A of Table 2.1 reports the summary statistics for the daily return sample of mutual fund
managers. This entire sample includes 4,185 mutual funds from 2001 to 2016. The sample of
mutual funds used for my fund ﬂows results are reported in 3.1. This second sample contains
7,928 unique fund series from 1,030 diﬀerent fund companies from 1995 to 2016.

I also run tests directly at the CRSP share-class level. I use individual share-classes to proxy for
diﬀerences in investor sophistication based on the separated retail and institutional sales channels.

12

All of the same data ﬁlters are applied to this second subsample. This sample is reported in Panel
B of Table 3.1.

For fund inﬂow and outﬂow data I use SEC ﬁling Form N-SAR from EDGAR. Since 2004,
mutual funds have been ﬁling Form NSAR semiannually with the SEC. Monthly inﬂows and
outﬂows are reported at the fund-series level in item 28. To ensure a consistent sample and
eliminate data reporting errors, I merge this data with my CRSP fund-series sample and require a
90% correlation for computed net ﬂows for the N-SAR ﬁlings and CRSP. This step retains a set of
2,320 unique fund series from 2004 to 2016.

1.6.2 Key Variables: Choice of Events

I construct a time series of national traumatic events and high-stress periods based on the daily
news programming listings from the Vanderbilt Television News Archive. Traumatic event days
are identiﬁed using a calculated measure based on the number of news outlets covering a story
along with the proportion of the nightly news coverage dedicated to the story.
I assume that
stories covered by the majority of news outlets for most of the broadcast period report to be highly
signiﬁcant events. I use keyword ﬁlters to eliminate events containing market-relevant information
such as politics and wars. This approach allows the aggregate news media to determine which
events are attention-grabbing and traumatic.

The Vanderbilt Television News Archive reports news broadcasts using a separate subject
heading for each segment of the evening broadcast. For example, the ﬁrst 5 topics for the January
17, 2003 broadcast of the NBC Nightly News is listed as: 1. Preview/Introduction Tom Brokaw
(New York); 2.
Iraq-US Relations / War Talk / Weapons Inspections; 3. Aﬃrmative Action /
University of Michigan / Rice; 4. Afghanistan / Accidental Bombing / Hearing; and 5. Stock Market
Report (Studio: Tom Brokaw). Data for other days and networks is presented similarly. Networks
have the discretion to devote as many of the evening broadcast segments to the same topic as they
desire. For instance, on April 16, 2007 the day of the tragic Virginia Tech shootings, the NBC news
broadcasts are listed as: 1. Virginia Tech / Massacre / Interviews; 2. Virginia Tech / Massacre;

13

Iraq War / Casualties; 4. Spring Weather / East Storm; and 5. Virginia Tech / Massacre /
3.
Steger Interview. On the same day, CBS is listed as broadcasting the following: 1. Virginia Tech /
Massacre / O’Dell Interview; 2. Virginia Tech / Massacre / Steger Interview; 3. Spring Weather /
East Storm; and 4. Virginia Tech / Massacre.

I calculate the proportion of the daily news broadcast devoted to a single story for each day in
the period from 1995 to 2016 across each network. To implement this step, I use the text prior to
the ‘/’ symbols in the segment description ﬁelds to categorize news topics as belonging to the same
story and calculate the proportion of the daily broadcast devoted to the same news story. I form an
initial list of news events from the pool of the top 10% most heavily covered news stories in the
sample period.

As a second step, I implement a word ﬁlter to eliminate news stories likely to contain market-
relevant information. Subject words relating to politics, wars, business, natural disasters (such as
Hurricane Katrina), and September 11th terrorist attacks are removed. This process retains a list of
63 unique major traumatic events between 1995 and 2016, by construction heavily reported across
the entire country. Figure 1.1 depicts the distribution of events over the sample period. The ﬁrst
portion of the sample period has a relatively uniform distribution of events, while the ﬁnal 3 years
of the sample experience a sharp rise in event frequency.

The list of events in my sample includes mass shootings and other attacks such as Columbine
Highschool, DC Sniper, Virginia Tech, Orlando Nightclub shootings, Boston Marathon Bombing,
Fort Hood Shootings, Aurora Colorado Movie Theater Shootings, and the Paris newspaper attacks.
Other events that received national attention, such as the Columbia space shuttle explosion, Princess
Diana’s death, and Michael Jackson’s death are also included.

As a ﬁnal step, to match the monthly frequency of my fund-ﬂow observations, I map the series
of daily events onto individual months. The intuition is that an event that occurs at the end of a
month would have minimal eﬀect on that month’s aggregate net ﬂows. For instance, if an event
occurred on the 28th of a given month, only the investment decisions made in the ﬁnal 2 days would
be aﬀected, whereas the investment decisions made during the previous 28 days of the same month

14

would be untreated. Furthermore, the stress eﬀects would carry over into the following month.

I designate event months using a simple rule. For each event, I create a four- week ‘event
window’ and proceed to sum for each month the number of ‘event days’ during that month. Months
with at least 20 ‘event days’ are considered as ‘event months.’ In other words, if the event occurred
at the beginning of the month, that month is counted as a treatment month. If an event occurred
at the end of the month, I count the following month as a treatment month. If the event happened
during the middle of the month, I do not count that month unless there is an additional nearby event.
Figure 1.2 illustrates these rules.

15

CHAPTER 2

FUND MANAGERS

2.1 Motivation and Methodology

In this study, I examine how stress aﬀects the decision-making abilities of presumably so-
phisticated mutual-fund managers. This setting has two key advantages. First, I can observe the
daily return performance of individual managers. Although I cannot directly observe the fund
managers’ actual decision-making process, daily returns are a direct outcome of the unobserved
decision-making process. Second, fund managers have a known geographic location. This allows
me to exploit exogenous variation in the geographic exposure of managers to each traumatic event
and thereby identify with reasonable precision the eﬀects of stress on managerial performance.

Fund managers have a pre-existing geographic distribution across population centers in the
United States (Pool et al. (2012)). Similarly, the locations of the traumatic events in my sample
are highly correlated with population centers across the country, with few events occurring in rural
areas. Figure 1.3 depicts the locations of traumatic events over a map of the United States. The
majority of events are located in coastal areas with high density populations. For each of these
events, managers in close geographic proximity to the event are more likely to have experienced
greater stress exposure (Gabriel, Ferrando, Cortón, Mingote, García-Camba, Liria, and Galea
(2007), Rubin, Brewin, Greenberg, Simpson, and Wessely (2005)).

The emergence of PTSD symptoms is related to a perceived threat to one’s personal integrity
(Neria, Nandi, and Galea (2008)).
Importantly, the threat to the individual need not be actual.
Geographic proximity to an event is positively related to increased knowledge and salience of the
traumatic event. Individuals are more likely to have visited the location or have personal familiarity
with it, creating pathways toward vividly imagining the event. Local media is also more likely to
cover heavily the event taking place within its coverage area, increasing media exposure to images
and videos of the event.

16

Therefore, fund managers in close proximity to an event are more likely to be aﬀected by stress
and post-traumatic stress symptoms than distant fund managers are. I treat my series of traumatic
events as a series of exogenous shocks to a geographic location. Fund managers local to an event
will be exposed to a relatively high level of post-traumatic stress compared to distant managers. As
a result, these local managers will face an increase to their marginal cost of information processing
and have greater diﬃculty in making investment decisions. For each event, I treat this local group
of managers as an exogenously deﬁned treatment group, while managers distant to the event are a
natural control group.

Using the manager oﬃce locations in CRSP, I measure the distance between each mutual fund
and event. I manually identify zip codes for each event, map the set of fund and event zip codes
onto latitude and longitude coordinates, and calculate the pairwise distances between events and
funds using the Haversine method. Managers whose oﬃces are within 100 miles of an event are
deﬁned to be the sample of ‘near’ managers, while those 1,000 miles from the event are designated
as the ‘far’ control group subsample.

I then separately examine ‘Foreign Equity’ and ‘Domestic Equity’ mutual funds.

‘Foreign
Equity’ funds are managed within the United States, but have overseas portfolio holdings. This em-
pirical design avoids any potential home bias (Pool et al. (2012)) or local information eﬀects (Shive
(2011)) that managers may have about ﬁrms in the local region surrounding their oﬃces. Neither
the ‘near’ nor the ‘far’ group of managers have a geographically-based information advantage when
investing in international ﬁrms. This feature helps isolate the information about the traumatic event
itself from the information about the cash ﬂows of portfolio companies. For example, information
about a US school shooting is almost entirely irrelevant for a mutual fund investing in the Middle
East. Any additional market-wide eﬀect of the traumatic event will aﬀect both the treatment and
control group managers. The cash ﬂows of foreign ﬁrms are unlikely to be aﬀected by local US
shocks, and changes to market sentiment will aﬀect near and distant funds equally.

Table 2.1 reports summary statistics for my sample of mutual fund managers. Panels B and C
display data separately for the Domestic and International equity mutual funds. International funds

17

tend to be smaller and younger than their domestic counterparts, and to have higher expense ratios.
International equity funds also typically have less turnover than domestic equity funds do.

To further control for potential geographic diﬀerences between the treatment and control man-
ager groups, I use propensity score matching to identify a set of treatment and control group funds.
The matching criteria I include are investment objective, past returns, fund size, fund age, and
portfolio turnover. I separately perform this matching step on both the domestic equity and inter-
national equity group of funds. This process ensures that the treatment and control group funds
will have similar distributions along these observable dimensions.

Table 2.2 reports subsample summary statistics for the near and far matched samples. Panel (A)
reports summary stats for the sample of 361 ‘near’ international equity funds with the corresponding
632 ‘far’ funds are shown in Panel B. As expected, the two subsamples are fairly consistent. Panels C
and D similarly show summary statistics for the ‘near’ and ‘far’ domestic equity manager samples,
for a total of 1,031 unique ‘near’ managers and 1,804 ‘far’ managers. Once again, these two
subsamples, by design, have highly similar observable fund characteristics.

I begin by deﬁning a one-quarter event window (63 trading days) before and after each traumatic
event in the sample. Taking advantage of the localized shocks to stress following traumatic events, I
employ a Diﬀerence-in-Diﬀerences approach to testing the eﬀects of stress on manager investment
abilities. The general speciﬁcation used to test hypotheses is:

Yit = β1Post_Eventi,t + β2Neari,t + β3Post_Eventi,t xNeari,t

+ Controlsi,t + Eventt + Objectivei + i,t

(2.1)

In this model, Y represents various fund characteristics such as risk factor weightings, turnover,
or benchmark tracking errors estimated over the pre-event and post-event window. The primary
variable of interest is the interaction between Post_Eventi,t and the near fund indicator Neari,t.
This variable estimates whether the average change in fund characteristics between the pre- and
post-event periods is conditional upon proximity to the event itself.

18

The above speciﬁcation includes controls for various fund characteristics, including fund size,
fund age, and turnover. Additionally, the speciﬁcation also includes event (time) eﬀects and
investment objective eﬀects.

As a ﬁnal step, to control for potential omitted time-series or geographic eﬀects, I perform a
series of simulations selecting 1,500 randomized event days and locations to create test statistics
for the purpose of testing against the observed pre- and post-event changes in fund characteristics.
Based on Figure 1.1, I draw treatment days from two separate uniform distributions for the 2001
to 2013 and 2014 to 2016 periods to capture the actual observed event frequency distribution. As a
second step, I randomly assign the actual traumatic events to each of the simulated treatment dates.
This randomizes the order in which events occur, but maintains the existing geographic distribution
of the events. This process generates the entire distribution of diﬀerences in fund characteristics
between ‘near’ and ‘far’ fund manager groups.

2.2 Fund Manager Results

2.2.1 Primary Results: Fund Returns

The mutual-fund industry is a competitive market and managers optimize information usage to
pursue proﬁtable investment opportunities. If there is a shock to the marginal cost of information
processing, managers will tend to use less information (and interpret whatever information they
do us with a greater error rate) in shaping investment decisions. To the extent that information
is beneﬁcial to managers, an increase in information costs will reduce investment performance,
resulting in lower realized returns for high stress managers.

Therefore, my ﬁrst hypothesis is: Hypothesis 1: Fund managers located in close proximity to

stressful events will underperform compared to geographically distant peers.

To test this hypothesis I implement an event study approach. First, I construct an event window
(t−21 to t+63 surrounding each traumatic event in my sample) and calculate event period returns
for both the ‘near’ and the ‘far’ fund manager groups. My ﬁrst set of results directly benchmarks
the ‘near’ group of manager returns against the ‘far’ group. I then repeat the same procedure, but

19

instead compare Fama-French 3 Factor returns between the geographic groups.

Figure 2.1 reports the initial results. The green line depicts the cumulative diﬀerence in the
matched sample of ‘near’ and ‘far’ international equity mutual funds. The cumulative pre-event
diﬀerence in returns is relatively small and is statistically insigniﬁcant. In the course of the ﬁrst
month of trading, day 0 to day 21, there is a gradual decline in returns from -10 basis points to -25
basis points. During the second month of trading, the underperformances intensiﬁes to -60 basis
points. By the third month, there is no additional underperformance, with the two sets of funds
having similar performance.

The observed delay in eﬀect, with minor underperformance over the initial trading month is
not consistent with a distraction-based interpretation. While managers may be distracted for a
short period of time surrounding the very occurrence of an event, it is unlikely that this would
signiﬁcantly aﬀect returns 4-8 weeks after the initial period of distraction.

To assess whether this underperformance is an artifact of the particular event days chosen, I
perform a series of simulations to create test statistics for the stress eﬀect.1. Figure 2.2 plots
the median and bottom 5th percentile of simulated returns along with the actual observed return
diﬀerence for the event period window. Results for international equity and domestic equity funds
are separately reported in Panels A and B. As expected, there is a negligible performance diﬀerence
at the median between the matched groups of mutual-fund managers. The cumulative diﬀerence in
returns 8 weeks after the traumatic event is signiﬁcant for both manager subsamples. The pattern
of underperformance for domestic equity managers closely reﬂects that of the international group.
The performance of local managers gradually declines over the course of the ﬁrst month following
the event, and the underperformance increases between 4 and 8 weeks.

In Figure 2.3 I repeat the above analysis using abnormal returns calculated with the Fama
French 3 Factor Model. I use the 90 days prior to each event window to estimate factor loadings
and calculate abnormal returns over the subsequent period. This approach further controls for any
1I also calculate cumulative abnormal returns variances over the 60 day event window using
the event study methodology from MacKinlay (1997). Results from this approach are qualitatively
similar and provide consistent statistical inference.

20

additional potential diﬀerence in expected returns between the ‘near’ and ‘far’ groups not captured
by the previous matching process. Results are qualitatively similar; the results, depicted in Figure
2.2, show small declines over the ﬁrst month following the event with rapid declines 4-8 weeks
post-event.

Table 2.3 reports the actual cumulative abnormal returns along with the left tail of the simulated
return distribution. Panel A shows the results for the group of international equity fund managers.
The second column displays the actual event returns in basis points for each week of the event
window. Columns (3) to (5) list the left tail of the simulated returns from randomized event days.
The simulated median return is in the ﬁnal, 6th column. The underperformance for domestic equity
managers over the ﬁrst 30 days is insigniﬁcant. Whereas some attention or distraction eﬀects could
be related to the traumatic events, these eﬀects do not appear to have a materially stronger on local
managers. After the ﬁrst 30 days, there is a mostly monotonic decrease in cumulative abnormal
returns for domestic equity managers up to one quarter following the event. This performance gap
is signiﬁcant at the 5% level. Panel (B) of Table 3.7 reports a similar pattern of results for domestic
equity funds.

Simulated results using abnormal returns are shown in Table 2.4. The international equity
results in Panel (A) are consistent with those reported in Table 2.3. Over the ﬁrst month post-event,
the diﬀerence in cumulative abnormal returns is approximately zero. Over the next two months,
local funds underperform their distant counterparts by 80 basis points. This result is signiﬁcant at
the 5% simulated level. Results are slightly diﬀerent with abnormal domestic equity returns, with
a gradual decline of 20 basis points over the initial month.

2.2.2 Additional Results

2.2.2.1 Risk Exposure

Besides stress, one alternative explanation for the underperformance of local fund managers is a
change in post-event expected returns. Wang, Albert and Young, Michael (2018) ﬁnd that after

21

terrorist attacks mutual-fund investors increase their allocations toward less risky asset classes;
they interpret this ﬁnding as a relative increase in risk aversion following terrorist activity.
If
mutual-fund managers become more fearful after traumatic events, potentially through changing
perceptions about the likelihood of disaster events, their portfolio allocations may become less
risky and the observed underperformance may be driven by an intentional reduction in risk-factor
loadings rather than impaired decision-making abilities. Antoniou et al. (2018) similarly use
geographic distance between ﬁnancial analysts and terrorist attacks to measure the relative eﬀect
of the terrorist attack on analyst risk aversion.

My second hypothesis is: Hypothesis 2: Traumatic events are a greater shock to risk aversion

for fund managers in close proximity to the event.

I test this hypothesis by estimating the Fama French 3 factor and momentum factor weightings
for each fund over the 63 trading days prior to each traumatic event and, again, for the 63 trading
days following each event.
I use a diﬀerence-in-diﬀerences speciﬁcation while controlling for
turnover and pre-event risk factor weightings to test whether there is an overall change in risk-factor
exposures following traumatic events.

The results of this test are reported in Table 2.5. Column (1) of Panel A shows the change in
market betas after traumatic events. I ﬁnd a decline in market betas across the entire sample of
domestic-equity funds. However, because the interaction term Post_EventxNear is insigniﬁcant,
I ﬁnd no evidence that funds in close proximity to traumatic events experience an additional decline
in betas. Columns 2-4 report results for changes in Small-Minus-Big, High-minus-Low and the
Momentum factors.
I ﬁnd no evidence of a change in any of these additional risk factors for
domestic equity funds.

Panel B of Table 2.5 shows the same results for my sample of International Equity funds.
These results are largely consistent with those of the domestic equity funds, with little evidence of
changes in risk exposures between near and far funds following traumatic events.

Panel C displays the simulation results for the change in market-risk exposure following trau-
matic events. The observed actual diﬀerences in market betas are approximately equal to the overall

22

simulated median from selecting random event days from the sample. There is little evidence that
mutual-fund managers adjust their portfolio holdings in a manner consistent with an increase in risk
aversion following traumatic events. A change in expected returns prompted by shifting portfolio
weights is unlikely to explain the observed underperformance by local funds.

2.2.2.2 Fund Turnover

The symptoms of post-traumatic stress disorder oﬀer two potential explanations for the observed
underperformance of local mutual funds in the aftermath of traumatic events. In an environment in
which with increased marginal information processing costs could prompt the managers to respond
by either applying mental eﬀorts toward fewer investment decisions or by reducing the investment
opportunity set that they will consider. Either of these options would reduce the cognitive burden
of the investment decision-making process and, conceivably, reduce relative portfolio performance.
In considering the ﬁrst explanation, if managers reduce their intensity of trading activity, on
average, fewer proﬁtable investments will be made and the remaining portfolio positions may grow
stale, reducing the daily fund returns observed in the aftermath of traumatic events. Managers
self-select whether to pursue a high or low investment turnover strategy. Within the subset of
managers pursuing a high turnover strategy, a shock to marginal information processing costs is
more likely to be a binding constraint on the number of trades made, resulting, at the same time,
in a larger reduction in fund turnover. Similarly, low-turnover funds will likely see a lesser shift
in turnover between the pre- and post-event period because managers had previously self-selected
into investment strategies where information processing costs were not a binding constraint on the
number of trades being made.

My third hypothesis is High stress managers in close proximity to traumatic events will reduce
their investment turnover compared to distant managers. This eﬀect will be more pronounced for
managers pursuing a high-turnover strategy prior to the event.

Empirically, the quantity of fund-trading activity is observable through portfolio turnover,
provided on a quarterly basis in the CRSP database. This variable is calculated as the dollar value

23

of the prior 12 month purchases and sales divided by the average total assets of the fund over the
prior 12 months. For each fund in my sample, I classify funds in the bottom quartile of turnover as
‘low turnover’ and funds in the upper quartile of turnover as ‘high turnover’. Because turnover is
reported at the quarterly frequency, I use the nearest observation before and after each endogenously
occurring event. This introduces some noise into my turnover measurements because turnover may
be reported between one and three months following each event. However, this measurement error
noise is unlikely to bias my results in favor of ﬁnding results.

Turnover results are reported in Table 2.6. Columns 1 and 2 display results separately for the
low- and high-turnover groups. Column 3 shows results for the diﬀerence between the low and
high turnover subsamples. For this test, a joint model is estimated with the full set of interactions.
The diﬀerence between coeﬃcients will have a χ2 distribution. Panel A reports the results for
the set of domestic-equity managers. The variable of interest, Post_EventxNear, is negative and
signiﬁcant for both the low- and high-turnover funds. The point estimate for high turnover funds
is −0.88. This constitutes a 24% annual reduction in asset turnover for funds in close proximity to
traumatic events.

Panel B of Table 2.6 shows results for the international fund sub-sample. In that subsample,
high-turnover managers near traumatic events once again decrease their turnover relative to distant
funds. The diﬀerence between low- and high-turnover funds is statistically signiﬁcant. For high-
turnover international funds, the estimated decrease in turnover for Near funds is −.21, a 15%
reduction in annual turnover.

Panel C displays simulated results for the near and far funds. For international funds, managers
in close proximity to traumatic events have 20% lower turnover post-event. This diﬀerence is
signiﬁcant at the 95th simulated percentile. Distant managers have no decrease in turnover.
Similarly, for domestic equity managers, ‘near’ managers have a 40% decline in turnover following
events, also signiﬁcant at the 95th percentile. Distant managers have 4.5% decline in turnover, but
this decline is not signiﬁcant.

24

2.2.2.3 Fund Indexing Strategies

For the second explanation, I consider whether the manager’s investment strategy changes in the
aftermath of traumatic events. Within a given investment objective, managers must strike a trade-
oﬀ between the extent to which they will replicate their benchmark index (‘closet indexing’) and
the extent to which they will actively deviate from the index (‘active management’) (Cremers and
Petajisto (2009)). From this perspective, replicating the underlying index is the default position for
a manager because the requisite portfolio weights are known in advance and require minimal eﬀort
to determine. In contrast, active management is a more cognitively intense strategy because the
portfolio positions are determined through a process of ﬁnancial research and analysis. If managers
local to a venue of a traumatic event have skill in selecting securities, forgoing proﬁtable active-
management opportunities to pursue an increased indexing strategy will lead to their observed
underperformance. This is intuitive because the only way to outperform a given benchmark is
intentionally to deviate from the benchmark security weights.

For managers who have chosen to follow a strategy of closely tracking their benchmark, when
making a investment allocation decision the ‘default’ option is to invest in the benchmark itself.
When faced with a shock to their level of stress, these managers are more likely to maintain the
status quo and invest in the benchmark. However, for the set of managers pursuing an active strategy
who have made large deviations from their benchmark, it is less clear what these managers would
consider to be their ‘default’ option as they have explicitly chosen not to track their benchmark
closely. For managers who have chosen to follow a strategy of closely tracking their benchmark,
the ‘default’ investment-allocation decision s to invest in the benchmark itself. When faced with a
shock to their stress level, these managers are more likely to maintain the status quo and continue
investing in the benchmark. However, the ‘default’ option for the set of managers pursuing an active
strategy, involving large deviations from their benchmark, is less clear.

My fourth hypothesis is Managers pursuing a ‘closet indexing’ strategy will increase their
weights in their respective index in the aftermath of traumatic events. This eﬀect will be greater for
mangers with exposure to stressful events.

25

Empirically, this can be observed by measuring tracking errors between managers and their
self-selected benchmark. Following Cremers and Petajisto (2009)), for each event I identify the
benchmark index whose returns most closely align with the actively-managed fund by using the
following test:

ri,t = α + βrindex,t + t

(2.2)

At the start of each calendar year, I estimate this equation using the full set of U.S. equity benchmark
returns from the St. Louis Fed data website (FRED). 2 For each fund, I select the index whose
estimated model has the highest R2 as the benchmark index for that calendar year.

After the benchmark indices are identiﬁed, I proceed to measure event-time tracking errors
for the ‘near’ and ‘far’ subsamples. Tracking errors are deﬁned as the standard deviation of the
diﬀerence in returns between the fund and underlying benchmark. I categorize funds in the bottom
quartile of tracking errors as ‘closet index’ funds and funds in the upper quartile as ‘active’ funds.
To remove potential noise from the event itself, I drop the initial 21 days before and after each event
and calculate tracking errors over the remaining 42 days pre- and post-event.

Table 2.7 reports the results of this test. The variable of interest is the interaction between
post-event and the indicator for near funds. Column 1 displays results for the set of closet-indexing
funds. The interaction term has an estimated coeﬃcient of −0.08, a 12% reduction in post-event
tracking errors for the group of managers near events. On the other hand, active fund managers
increase their tracking errors by 32%. This increase in tracking errors could either be due to an
intentional increase in active positions or a result of stale portfolio positions.

2.3 Alternative Hypotheses

The eﬀects I document could be driven by a pure investor attention or distraction eﬀect (Hirsh-
leifer, Lim, and Teoh, 2009). This would suggest that investors neglect ﬁnancial information while
2The major U.S. indices considered are: Russell 1000, Russell 1000 Value, Russell 2000,
Russell 2000 Growth, Russell 2000 Value, Russell 3000, Russell 3000 Value, Russell Midcap,
Russell Midcap Growth, Russell Midcap Value, S&P 500, Wilshire 4500, Wilshire 5000.

26

traumatic events are occurring because they choose to pay attention to the event instead. Under
this hypothesis, all of the eﬀect should occur during a short window after the event, while the event
is heavily covered by the news media, and with negligible lingering eﬀects over the next several
weeks. The time series of daily returns facilitates the measurement of when performance shifts
happen. Information shocks should aﬀect trades and decisions made immediately following the
event, rather than aﬀect medium to long term performance. Although I cannot directly observe
trades and decisions from the daily return data, delays in performance shifts after the event are
consistent with a stress hypothesis.

Based upon the timeline of event returns in Table 2.3 and Table 2.4, this distraction hypothesis
is unlikely to explain my results. The cumulative underperformance by local funds is negligible
over the ﬁrst 10-21 days post-event. If managers are temporarily distracted by the event, they would
quickly revert to prior abilities and subsequent trades would be unaﬀected. However, a shock to
stress will impair abilities over the following 8-12 weeks, consistent with the observed eﬀect.

Additionally, the eﬀects may be driven by a change in investor sentiment. Traumatic events may
change expectations of future returns and cash ﬂows and alter incentives to invest in mutual funds
(Ben-Rephael, Kandel, and Wohl, 2010). By restricting my sample to international equity funds for
my tests, I am able to rule out this hypothesis. These are funds headquartered in the United Sates
investing in overseas companies. This controls for home bias and any local information eﬀects for
the managers. Furthermore, any information about the traumatic event is especially irrelevant to
foreign equity portfolio returns. If traumatic events aﬀect the market as a whole and shift investor
sentiment, both the ‘near’ and ‘far’ groups of funds will be aﬀected and any changes will be netted
out in the diﬀerences between the two groups and event eﬀects.

27

CHAPTER 3

MUTUAL FUND CLIENTS

3.0.1 Methodology

In my next set of results, I examine how stress aﬀects the clients of mutual funds in their decisions
to allocate wealth across the mutual-fund sector. Mutual-fund ﬂows represent the outcomes of
investment decision-making processes by individual investors. While I am unable to observe daily
fund ﬂows, I can estimate changes to the cross-section of fund allocations across time. This enables
the measurement oﬀ the sensitivity of net ﬂows to information about observable fund characteristics.
This setting complements that of the mutual-fund managers and fosters a deeper examination
into the choices made by individual investors. If stress increases the marginal cost of information,
investors will demand less information. A study of post-event mutual-fund ﬂows allows for a more
precise examination of how information is used to inform investment decisions and how these
decisions are aﬀected by a shock to relative information costs.

To test my hypothesis, I ﬁrst calculate fund ﬂows using the standard approach in the literature.
Following Sensoy (2009), for each fund I back out net fund ﬂows from the proportional diﬀerence
in monthly total net assets less the returns earned on fund assets.

Flowi,t,t+1 =

T N Ai,t+1
T N Ai,t

− (1 + Ri,t+1)

(3.1)

Next, I specify a baseline empirical model to estimate expected fund ﬂows to fund i for time
i for time t to t + 1. I include in this speciﬁcation a set of indicator variables specifying the past
return quintile of fund i. Each month, I rank funds by their past year and quarter return within
their investment style group and assign each fund to a past return quintile. This set of past return
indicators is intended to capture the convexity of the ﬂow-performance sensitivity (Chevalier and
Ellison (1997), Sirri and Tufano (1998), Lynch and Musto (2003),Jonathan B. Berk and Richard C.
Green (2004)). I also include various fund characteristic controls (such as fund age, fund size, and

28

expense ratio), time eﬀects, and investment style eﬀects based on CRSP objective codes.

Flowt+1,i = α + βReturn_Quintilet,i + βSearch_Costs

+ Controlst,i + Y eart + Fund_st ylei + t,i

(3.2)

I estimate this model separately for both normal and high stress periods. Standard errors are
clustered at the fund level. I employ a methodology similar to a Chow Test to compare coeﬃcients
between the normal and high stress periods. First, I estimate a combined model incorporating a full
set of interaction terms for both periods. For each coeﬃcient, I then estimate the model under the
restriction of no diﬀerence between normal and stress periods. Diﬀerences in residuals between
the restricted and unrestricted model follow a χ2 distribution with n degrees of freedom.

3.1 Mutual Fund Client Results

3.1.1 Fund Visibility

My ﬁfth hypothesis examines how stress aﬀects the formation of investor choice sets. Odean (1999)
suggests that, when making an asset-allocation decision, investors ﬁrst form a set of securities to
consider and then decide within this reduced set of investment opportunities. Relatively small and
unknown funds are more diﬃcult for investors to consider because learning about them requires
more eﬀort. During periods of high stress, the cognitive eﬀort required to seek out additional funds
will be more costly to investors. As a result, cognitively-constrained investors will be less willing
to exert the eﬀort required to ﬁnd smaller, less known funds and will be more likely to limit their
choice set to highly visible large funds.

This leads to my ﬁfth hypothesis: Hypothesis 5: In periods of high post-traumatic stress and
reduced investor concentration, funds with lower search costs and greater marketing expenditures
will receive a greater portion of fund ﬂows relative to normal periods

Following Sirri and Tufano (1998), I proxy for search costs using fund size and 12b1 fees.
Under this hypothesis, the proportion of funds allocated to both larger ﬁnds (more monthly assets

29

under management) and funds with increased marketing and sales commission costs (12b1 fees)
should be greater under stress periods than under normal periods. Table 3.2reports these results
for the universe of CRSP fund series from 1995 to 2016. Panel A displays results using rolling
12-month past returns to assign return quintiles while Panel B uses past quarter returns.

Column (1) shows results under normal (non-stress) time periods. The coeﬃcient associated
with log(Monthl y_Total_Net_Assets) in Panel A is −0.0093. The negative estimate is consistent
with the literature because larger funds, simply by virtue of their asset size, generally have decreasing
fund growth rates. During periods of high stress, reported in Column (2), this coeﬃcient is less
negative, with an estimate of −0.0063. This increase between time periods of −.003 is highly
statistically signiﬁcant. At the 90th percentile of fund size, this diﬀerence between normal and
stress periods is equivalent to a 1% increase in expected fund ﬂows. The proportion of fund ﬂows
allocated to large funds increases following traumatic events, consistent with the interpretation of
an increase in investors’ marginal search costs. Results are consistent across Panels (A) and (B)
regardless of how past return groups are deﬁned.

Results for 12b1 fees follow a similar pattern. When using past year returns (Panel A), the
coeﬃcient estimate for actual 12b1 fees is −1.5599; this increases to 0.6949 during the stress
periods. Once again, the diﬀerence between the two periods is highly statistically signiﬁcant and is
equivalent to a .2% i increase in expected fund ﬂows at the mean 12b1 fee level.

When information costs are low, investors are sensitive to mutual-fund fees, resulting in a
negative coeﬃcient estimate for the normal periods. However, this sensitivity is reduced during
high-stress periods. 12b1 fees are predominately associated with broker-sold funds (Yang, 2016).
This increase suggests an increase in ﬂow towards broker-sold funds during high-stress periods. 1
These estimated results are consistent with Hypothesis 5. When stress-related search costs are
high, the distribution of fund ﬂows shifts towards funds with low search costs, that is, large funds
with high marketing-related fees.

1Two explanations could be that investors are more willing to pay for broker advice during these

periods or brokers are able to exploit less sensitive investors and sell more expensive funds.

30

3.1.2 Flow-Performance Sensitivity

In normal environments, investors chase past performance and allocate wealth towards mutual
funds that have outperformed in the prior period (Chevalier and Ellison (1997), Sirri and Tufano
(1998), Lynch and Musto (2003),Jonathan B. Berk and Richard C. Green (2004)). The cognitive
impairments to concentration and learning ability that stem from shocks to stress limit the amount
of information, including information concerning past returns, that can be processed before the
investor feels overwhelmed. Investors may still weigh past returns heavily in their decisions, but
the ability to identify funds with their preferred ﬁnancial characteristics is reduced. Empirically,
this can be estimated by separately measuring the ﬂow-performance convexity for normal and
stress-related periods.

Therefore, my sixth hypothesis can be stated as: Hypothesis 6: Following traumatic event
shocks, the sensitivity of fund-ﬂow response to fund information is reduced relative to normal time
periods

Figure 3.1 depicts these results using return deciles. This ﬁgure was estimated using past 12
month return rankings. The red line represents the familiar ﬂow-performance sensitivity under
normal circumstances. The top-performing funds receive approximately 4% expected monthly net
ﬂows. This quickly drops oﬀ, with the 9th decile earning 2.1% in expected net ﬂows. Subsequent
deciles decline approximately linearly, with the bottom decile receiving only about .5% expected net
ﬂows. The ﬂow-performance sensitivity materially changes during periods of high post-traumatic
stress; it becomes ﬂatter. The expected ﬂows to the highest-performing funds are almost halved,
with the top-performing funds only receiving 2.5% higher net ﬂows following traumatic events
compared to 4% for the normal period. The lower deciles also display less sensitivity with a
ﬂattened slope. Both the 9th decile and the bottom decile both have the same 1.75% estimated
expected net ﬂows. This evidence suggests that investors weigh past return information less heavily
during high stress periods.

Table 3.3 reports the estimates for the ﬂow-sensitivity tests. Panel (A) reports results using the
prior 12 month returns to construct return quintiles. There is a monotonic increase in net ﬂows

31

during the normal period between the bottom quintile and the top quintile. The ﬂow-performance
relation is highly convex. The top-performing funds receive 2.26% more expected net ﬂows than
the 3rd quintile funds do. This increase, directed toward the top past performers, is four times
greater than the range between the ﬁrst and 4th quintiles (.5%). Column (2) shows that this
convexity decreases following traumatic events. Flows directed to the top quintile during stress
periods are almost halved, with the top quintile only receiving 1.3% higher expected net ﬂows.
The diﬀerence between normal and stress periods, reported in Column (3), is highly statistically
signiﬁcant. Furthermore, these results carry over to Panel (B), characterized by the use of past
quarter returns. Once again, during high-stress periods, the proportion of ﬂows allocated to the
highest-performing funds is almost halved compared to the normal periods.

The pattern of reduced sensitivity to ﬁnancial characteristics carries over to other aspects of
mutual funds. The coeﬃcients on both expense ratios and return risk (volatility) are smaller in
magnitude, indicating a reduced sensitivity of ﬂows to fund fees and fund risk. These results are
consistent across Panels (A) and (B). However, interpreting coeﬃcients is less clear in this setting
because the size of a fund family may be related to search costs or fund spillover eﬀects Massa
(2003).

These results are consistent with Hypothesis 6. This evidence suggests a ﬂatter slope for the
ﬂow-performance relation during high-stress periods. When stress is high and individuals have less
ability to concentrate and acquire new information, ﬁnancial decision outcomes are less sensitive
to relevant ﬁnancial information.

3.1.3 Retail and Institutional Share-Classes

Expertise with analyzing ﬁnancial information may mitigate the eﬀects of stress on ﬁnancial
decision-making processes. Retail investors with low ﬁnancial sophistication face signiﬁcant
barriers in accessing ﬁnancial information. This group of low sophistication investors accesses
mutual fund information through channels such as brokers, mutual fund prospectus statements,
or premium subscriptions to online websites such as Morningstar. Post-Traumatic Stress also

32

reduces individuals’ ability to learn new information. For retail investors, who already have high
marginal costs of gathering ﬁnancial information, the increases in information processing costs
from post-traumatic stress will likely severely limit the amount of information incorporated into
their investment decisions. On the other hand, sophisticated institutional investors have specialized
tools and processes to handle this type of information and likely have more robust decision-making
structures in place. The information-gathering processes for high sophistication investors is less
likely to be directly aﬀected by the psychological eﬀects of stress.

My seventh hypothesis is then: Hypothesis 7: The eﬀects of stress on decision-making will be
more pronounced among less sophisticated investors than among sophisticated investors with high
information-processing ability

To test this hypothesis, I perform the same ﬂow-sensitivity tests using separate subsamples of
retail and institutional share classes to proxy for investor sophistication. My results are reported in
Table 3.4 where Panel (A) reports results for the retail subsample, and Panel (B) reports results for
institutional subsample. The same set of controls, eﬀects, and standard errors are used as in Table
3.3.

For retail investors, the results show a similar pattern as in Table 3.3. The normal period results
depict the familiar convex ﬂow-performance sensitivity, with the top-performing funds receiving
1.66% higher net ﬂows than the 3rd quintile of funds. During high-stress periods, the proportion of
ﬂows to the highest-performing funds drops to .91%. The share of ﬂows going to the 4th quintile
of funds also decreases signiﬁcantly. Similar to Panel (B) of Table 3.3, the worst-performing funds
have a small increase in expected ﬂows, further suggesting that investors may pay less attention to
return performance.

Among institutional investors, there is no diﬀerence in the ﬂow-performance sensitivity between
normal and high stress periods for the top two quintiles of past returns. During normal periods,
the top quintile of funds receives 1.32% higher ﬂows. During high stress periods, this is reduced
to 1.17%. The diﬀerence between the two is not statistically signiﬁcant. Bottom-performing funds
receive slightly smaller ﬂows, but this is of marginal signiﬁcance.

33

Overall, these results are consistent with an interpretation of the shift in ﬂow-performance
sensitivity being driven by retail investors, who face barriers in accessing past return information.
Institutions with few barriers in researching mutual funds’ past returns display no change in behavior
that could be tied to their usage of return information.

3.1.4 Decision Making Ability

My next hypothesis addresses whether the investment decision-making ability is reduced by stress.
If information is beneﬁcial towards optimal asset allocation, increased information costs will lead
to a lower quality of observed decisions and investment alphas. While the literature has reported
mixed evidence as to whether the expected alpha of investor mutual funds decisions is positive
( Zheng (1999), Frazzini and Lamont (2008), and Akbas, Armstrong, Sorescu, and Subrahmanyam
(2015)), under stress and with increased information costs, there likely is a decrease in expected
alphas compared to normal periods.

My eighth hypothesis is then: Hypothesis 8: The portfolio of mutual funds chosen by investors
during high-stress periods underperforms relative to the portfolio of funds chosen by investors
during normal periods

To test this hypothesis, I construct a time series of ﬂow-weighted portfolios pursuing a methodol-
ogy similar to Zheng (1999). I create a series of mutual-fund portfolios weighted by the proportion
of net ﬂows each fund received in a given month. For each month, I identify funds with net ﬂows
greater than (and less than) 1.5 standard deviations above (below) their mean fund ﬂow. Investors
as a whole are taking an implicit long position in funds receiving a large portion of inﬂows, and
short position in funds with large outﬂows. I hold this monthly portfolio of net mutual-fund in-
vestment decisions for 24 months and calculate its 4-factor alpha (Carhart, 1997). This alpha is
representative of the investors’ mutual-fund selection ability as it increases when either the funds
that received large inﬂows outperformed on a risk-adjusted basis, or the funds that experienced
large outﬂows underperform. This can be interpreted as investors’ ability to improve their portfolio
allocations by shifting capital from underperforming funds to better mutual funds.

34

Panel A in Table 3.5 reports means and t-tests on the set of calculated net fund ﬂow portfolio
alphas. Both retail and institutional investors have an observed aggregate ability to improve their
portfolios by shifting capital into better-performing mutual funds.
I ﬁnd that, during normal
periods, the funds that have received large inﬂows outperform the funds that had large outﬂows by
12 basis points per month. Similarly, funds chosen by institutional investors outperform the sold
funds by 8 basis points per month. Following stress events, the ability of investors to shift capital
into higher-performing funds is impaired. Neither group of investors is able to earn a signiﬁcantly
positive alpha from their net fund ﬂow investment decisions. The aggregate portfolio of funds
purchased does not outperform the aggregate portfolio of funds sold.

The third column of Panel A reports the diﬀerences in alphas between normal and stress periods
and the joint hypothesis test that normal and stress period portfolio alphas are equal. For retail
investors, the decision-making impairment of 10 basis points is nearing conventional levels of
statistical signiﬁcance, with a t-statistic of 1.5. This is not surprising because a common investment
strategy, as evidenced by Table 3.4, is to chase past returns. A pure strategy of chasing past returns
is not likely to generate a large alpha or make eﬃcient use of ﬁnancial information.

Panel B reports the simulated distribution of net portfolio alphas using 3,000 draws of random-
ized event days. For each draw of the simulation, I draw random event days from two separate
uniform distributions, for the period from 1995 to 2013 and the period from 2013 to 2016, to mimic
the observed actual frequency of events during this period from Figure 1.1. I then calculate the
net high and low fund ﬂow portfolios and resulting alphas using the same process, and compare
the generated alphas using the actual stress event dates to the distribution of net alphas.2. Again, I
ﬁnd no evidence of an impairment to expected returns for the group of retail investors. While retail
investors may have less ability to pursue a strategy of chasing past mutual-fund performance it is
not clear that the expected return on their investments is signiﬁcantly decreased.

evidence of a diminished investment ability during periods of high stress.

However, for institutional investors, likely eﬃcient users of ﬁnancial information, I do ﬁnd
In Panel A, the 17
2This simulation approach is designed to alleviate concerns of my results being driven by
sampling bias. Standard time series regression methodology produces qualitatively similar results

35

basis-point drop in expected net alphas is highly signiﬁcant with a t-statistic of 2.13. Furthermore,
the decrease in portfolio alpha is at the far right tail of the simulated distribution (95% level),
suggesting that this observed decrease is unlikely to engender sample bias in the event days. When
information costs are high during stressful periods, institutions are unable to allocate as eﬃciently,
reducing their expected alpha by 17 basis points. While there is no evidence in Table 3.4 that
institutions use past return information diﬀerently between normal and high stress periods, the
increase in information costs from stress is likely aﬀecting the more sophisticated information
analysis processes that occur when determining institutional asset allocations.

3.1.5 Share Purchases and Redemptions

Finally, I examine whether mutual-fund share purchases or redemptions are more sensitive to
the cognitive eﬀects of stress. Purchasing new mutual fund shares is a cognitively diﬃcult task
that requires investors to identify investment opportunities from the entire universe of mutual
funds. Common PTSD coping mechanisms include ‘giving up’ or ‘avoidance.’ This increases the
likelihood that investors may choose to delay cognitively diﬃcult investment decisions. However,
redeeming shares is an easier decision to make because it only requires investors to consider the
smaller set of funds or assets in their personal portfolio. Furthermore, investors may not have the
ability to delay share redemption decisions if the decision is motivated by liquidity constraints or
tax-loss selling (Barber and Odean, 2000).

For these reasons, my ninth hypothesis is: Mutual fund share purchases will display greater

sensitivity to stress than share redemptions.

Results using separated inﬂows and outﬂows for my mutual fund sample are reported in Table
3.6. After controlling for past returns and fund characteristics, during normal periods the conditional
expected fund inﬂow reported in Column (1) is 8.2% of monthly total net assets. The expected
outﬂow during this sample period, reported in Column (2), is 5.2%. The Stress Ind coeﬃcient
reports the shift in expected inﬂows and outﬂows.
Inﬂows are reduced by .3%, while there is
no expected change in fund redemptions. This suggests that the change in net-ﬂow behaviors is

36

primarily driven through a reduction in mutual-fund purchases, rather than individuals pulling their
wealth out of the mutual-fund sector.

As a robustness check on the Form N-SAR sample, Column (3) repeats the same tests for net
in f lows − out f lows. Estimated coeﬃcients here are consistent with those reported in previous
tables.

3.2 Robustness Checks

I perform two out-of-sample robustness tests. In the ﬁrst test, I use my original set of mutual-
fund ﬂows, but change the event sample and only consider the out-of-sample September 11th
terrorist attacks and natural disasters. In the second test, I separately test the early and late time
periods of my sample to ensure the results are not driven by changes in investment patterns over
time.

In this test, I use the following 6 traumatic events:

In the ﬁrst test, I examine initially discarded traumatic events with my original sample of
fund ﬂows.
the September 11th terrorist
attacks, Hurricanes Rita, Katrina, and Sandy, and the Haiti and Tokyo earthquakes. These events
were initially discarded during my initial ﬁltering process because of the concerns for changes in
expected market cash ﬂows. While there might be some cash-ﬂow implications, these events are
also highly traumatic and directly associated with high levels of stress, anxiety, and PTSD-related
symptoms.

Table 3.8 reports the ﬂow-performance sensitivity results for the out-of-sample events test,
using the same methodology and data as in Table 3.3. Within this separate set of events and time
periods, the highest-performing funds receive a reduced proportion of fund inﬂows during periods
of elevated stress symptoms. These results are highly consistent with the previously reported
results.

Table 3.9 reports the second robustness test. I separately estimate the fund-ﬂow sensitivities
for the early and late sample periods using a set of interaction terms to identify high stress periods.
Column (1) limits the sample to 1995-2008, while column (2) uses 2009-2016. Consistent with

37

the original tests from Table 3.3, both periods estimate a similar reduction in ﬂow-performance
sensitivity for the highest quintile of fund returns.

38

APPENDICES

39

APPENDIX A

INTRODUCTION

40

Figure 1.1: The frequency distribution of events by year

This ﬁgure reports the number of unique traumatic events include in my sample set for each year
between 1995 and 2016. Events are chosen in a two part process. The ﬁrst step identiﬁes the top
10% of days with abnormally high news coverage of a single topic from the Vanderbilt Television
News Archive. The second step ﬁlters out these high coverage news days to discard events relating
to wars, politics, economics, business, natural disasters, and September 11th to limit the set of
events to identify events with limited market information.

41

0.02.55.07.510.012.519952000200520102015yearFirst.DayFigure 1.2: Event Month Formation Rules

Panel A: Beginning of Month

For each event, create 4 week window of treatment days and count months where two-thirds of the
month contain treatment days

Panel B: End of Month

Events at end of month are counted the following month. Overlapping windows are just counted as
single treatment days

42

Figure 1.3: US Event Locations

43

APPENDIX B

FUND MANAGERS

44

Figure 2.1: Cumulative Event Time Returns for International Equity Funds

This ﬁgure reports the event time cumulative returns for ‘near’ and ‘distant’ group of international
equity mutual fund managers across the sample of traumatic events from 2001 to 2016. The grey
line depicts the cumulative returns of the ‘near’ group of funds from 1 month(21 trading days)
prior to the event and the dashed orange line shows cumulative returns for the group of ‘far’ funds.
The triangle dotted blue line shows the cumulative diﬀerence in returns between the two sets of
managers from 1 month prior to the event up to 3 months(63 trading days) after the event.

45

−200−1000−200204060Event Time(Days)Cumulative Returns (Basis Points)Near FundsFar FundsDifferenceInternational Equity Funds Cumulative ReturnsFigure 2.2: Cumulative Near vs Distant Simulated Mutual Fund Performance

The following ﬁgures depict the cumulative diﬀerence in returns for the ‘near’ and ‘distant’ group
of managers across the sample of traumatic events from 2001 to 2016. The grey line depicts the
actual observed diﬀerence in returns between the matched sample of ‘near’ and ’far mutual fund
managers. Funds dare matched using a propensity score technique along observable characteristics
of investment objective, past returns, fund size, fund age, and portfolio turnover. 1,500 simulations
were conducted choosing randomized event days and locations. The orange line reports the median
cumulative diﬀerence in returns between the ‘near’ and ‘distant’ mutual fund groups. The blue line
reports the 95% percentile diﬀerence in simulated returns. Cumulative returns over the 1 month
prior to the event up to 3 months after the event are depicted. Panel A plots returns for the sample of
international equity fund managers and panel B plots the same returns for the sample of domestic
equity fund managers.

Panel A: Daily International Equity Mutual Fund Managers

46

−100−50050−200204060Event Time(Days)Cumulative Returns (Basis Points)Near FundsFar FundsDifferenceInternational Equity Funds Simulated ReturnsFigure 2.2 (cont’d): Cumulative Near vs Distant Simulated Mutual Fund Performance

Panel B: Daily Domestic Equity Mutual Fund Managers

47

−100−50050−200204060Event Time(Days)Cumulative Returns (Basis Points)Near FundsFar FundsDifferenceDomestic Equity Funds Simulated ReturnsFigure 2.3: Cumulative Near vs Distant Simulated Abnormal Mutual Fund Performance

The following ﬁgures depict the cumulative diﬀerence in abnormal returns for the ‘near’ and
‘distant’ group of managers across the sample of traumatic events from 2001 to 2016. Abnormal
returns are calculated as the daily residual return from the estimation of the Fama-French 3 factors
and an additional momentum factor over the quarter preceding the event window. The grey line
depicts the actual observed diﬀerence in returns between the matched sample of ‘near’ and ’far
mutual fund managers. Funds dare matched using a propensity score technique along observable
characteristics of investment objective, past returns, fund size, fund age, and portfolio turnover.
1,500 simulations were conducted choosing randomized event days and locations. The orange line
reports the median cumulative diﬀerence in returns between the ‘near’ and ‘distant’ mutual fund
groups. The blue line reports the 95% percentile diﬀerence in simulated returns. Cumulative
returns over the 1 month prior to the event up to 3 months after the event are depicted. Panel A plots
returns for the sample of international equity fund managers and panel B plots the same returns for
the sample of domestic equity fund managers.

Panel A: Daily International Equity Mutual Fund Managers

48

−100−50050−200204060Event Time(Days)Cumulative Returns (Basis Points)Near FundsFar FundsDifferenceInternational Equity Funds Abnormal Simulated ReturnsFigure 2.3 (cont’d): Cumulative Near vs Distant Simulated Abnormal Mutual Fund Performance

Panel B: Daily Domestic Equity Mutual Fund Managers

49

−100−50050−200204060Event Time(Days)Cumulative Returns (Basis Points)Near FundsFar FundsDifferenceDomestic Equity Funds Abnormal Simulated ReturnsTable 2.1: Summary Statistics

The following tables report summary statistics for the sample of mutual fund managers. Panel A
reports data for the entire sample of mutual funds from the CRSP universe aggregated to the fund
series level using a mapping of CRSP fundnos to EDGAR Series CIKs. This is the same mapping
as used for the accompanying fund-ﬂow tests. This data is for the 2001 to 2016 period when daily
data in CRSP becomes available. This data encompasses 4,185 distinct funds. Panel B reports the
subsample of Domestic Equity mutual funds. This includes 1,860 unique funds from 492 separate
fund companies. Panel C reports the subsample of International Equity funds in the CRSP dataset
which includes 646 funds from 252 fund companies.

Panel A: CRSP Universe Mutual Funds

Daily Return
Fund Size
Expense Ratio
12b1 Fees
Management Fees
Turnover
Fund Age(months)

mean median
0.000
0.000
2392.211
432.900
0.009
0.009
0.000
0.001
0.006
0.006
0.897
0.440
140.000
168.385

sd
0.010
9400.794
0.010
0.009
0.004
28.008
132.119

P10
-0.009
55.600
0.002
0.000
0.002
0.030
45.000

P90
0.009
4702.700
0.015
0.004
0.010
1.680
307.000

Panel B: Domestic Mutual Funds

Daily Return
Fund Size
Expense Ratio
12b1 Fees
Management Fees
Turnover
Fund Age(months)

mean median
0.001
0.000
448.500
2350.272
0.010
0.011
0.000
0.001
0.007
0.007
0.530
0.865
169.648
133.000

sd
0.014
10060.029
0.005
0.002
0.003
2.448
147.532

P10
-0.014
48.700
0.003
0.000
0.002
0.110
45.000

P90
0.014
4574.100
0.016
0.004
0.010
1.580
312.000

Panel C: International Mutual Funds

Daily Return
Fund Size
Expense Ratio
12b1 Fees
Management Fees
Turnover
Fund Age(months)

mean median
0.000
0.001
404.200
2132.455
0.012
0.012
0.001
0.000
0.008
0.008
0.500
0.683
139.157
118.000

sd
0.013
7995.388
0.005
0.002
0.009
0.832
95.611

P10
-0.013
52.700
0.006
0.000
0.004
0.110
42.000

P90
0.013
3838.200
0.019
0.004
0.012
1.330
254.000

50

Table 2.2: Summary Statistics: Matched Sample

The following tables report summary statistics for the sample of matched mutual fund managers.
Funds dare matched using a propensity score technique along observable characteristics of invest-
ment objective, past returns, fund size, fund age, and portfolio turnover. Near funds are deﬁned as
those with a headquarters zip code within 100 miles of a traumatic event. Far funds are those with
a headquarters zipcode greater than 1,000 miles from a traumatic event. Panel A reports results
for 361 near international equity fund managers. Panel B reports data for the 632 far international
equity fund managers. Panel C reports data on the 1,031 near domestic fund managers. Finally in
panel D, I show summary stats on the 1,804 far domestic fund managers.

Panel A: Near International Managers

Daily Return
Fund Size
Expense Ratio
12b1 Fees
Management Fees
Turnover
Fund Age(months)

mean median
-0.004
0.001
377.550
2563.602
0.012
0.012
0.000
0.001
0.008
0.008
0.500
0.716
144.331
116.000

sd
0.011
11266.673
0.005
0.002
0.003
0.854
98.708

P10
-0.020
45.100
0.006
0.000
0.004
0.130
44.000

P90
0.007
4226.300
0.018
0.003
0.011
1.390
264.000

Panel B: Far International Managers

Daily Return
Fund Size
Expense Ratio
12b1 Fees
Management Fees
Turnover
Fund Age(months)

mean median
0.000
-0.002
2337.357
395.500
0.012
0.012
0.000
0.001
0.008
0.008
0.440
0.646
147.043
117.000

sd
0.010
9235.853
0.005
0.002
0.003
0.778
103.815

P10
-0.015
49.000
0.006
0.000
0.004
0.100
43.000

P90
0.010
4110.900
0.018
0.003
0.012
1.290
272.000

51

Table 2.2 (cont’d): Summary Statistics: Matched Sample

Panel C: Near Domestic Managers

Daily Return
Fund Size
Expense Ratio
12b1 Fees
Management Fees
Turnover
Fund Age(months)

mean median
-0.000
-0.003
2776.801
456.200
0.010
0.010
0.001
0.001
0.007
0.007
0.937
0.550
146.000
175.487

sd
0.013
14648.014
0.005
0.002
0.003
2.866
148.416

P10
-0.024
46.880
0.003
0.000
0.002
0.120
44.000

P90
0.010
4573.900
0.015
0.003
0.010
1.630
325.400

Panel D: Far Domestic Managers

Daily Return
Fund Size
Expense Ratio
12b1 Fees
Management Fees
Turnover
Fund Age(months)

mean median
0.001
-0.000
2604.336
483.450
0.010
0.010
0.000
0.001
0.007
0.007
0.874
0.490
146.000
176.722

sd
0.013
11573.472
0.005
0.002
0.003
2.718
147.110

P10
-0.013
47.200
0.003
0.000
0.002
0.100
47.000

P90
0.014
5232.450
0.015
0.003
0.010
1.520
324.000

52

Table 2.3: Relative Daily Performance of Local and Distant Mutual Funds Following Stress Events

The following tables report estimated cumulative returns and simulated returns for U.S. based equity
mutual fund managers following stressful events. Panel A reports results using a sample of 361
U.S. based foreign equity mutual funds located within 100 miles of stressful events bench-marked
against a matched sample of funds headquartered 1,000 miles from the event. Panel B reports
results for 1,031 domestic equity mutual fund managers located with 100 miles of the traumatic
events. Daily returns for both groups are benchmarked against a matched sample of distant mutual
funds. The matching criteria used is: CRSP 3 digit objective code, Monthly Total Net Assets,
Quarterly Returns, Fund Turnover, and Fund Age. For each event day within the event window,
the abnormal return is calculated as the mean abnormal return across each of the 20 events in the
sample. Cumulative Abnormal Return (CAR) is calculated as the cumulative product of event days
from t0 to time t and reported in basis points. Columns (3) to (6) report simulated cumulative
abnormal returns based on 1,500 random draws with randomized event days and randomized event
order taken to match the original distribution of events across the sample period.

Panel A: International Equity Funds (Near less Far bps)

Event_Time (days) Actual (bps) Simulated_03 Simulated_05 Simulated_10 Simulated_50
-0.325
-0.742
-1.052
-3.212
-3.913
-4.381
-6.676

-9.166
-21.596
-28.931
-41.064
-51.831
-61.528
-73.538

-8.088
-17.095
-26.513
-36.420
-45.492
-53.213
-65.770

-6.077
-13.248
-20.075
-28.794
-36.041
-41.900
-54.387

-0.937
9.785
11.913
-0.748
-31.769
-53.927
-56.038

1
5
10
21
31
42
63

Panel B: Domestic Equity Funds (Near less Far bps)

Event_Time (days) Actual (bps) Simulated_03 Simulated_05 Simulated_10 Simulated_50
0.053
-0.075
-0.016
-0.411
-0.339
-0.965
-1.500

-4.818
-10.988
-14.920
-24.342
-28.905
-36.143
-41.938

14.389
8.566
0.247
-3.186
-12.074
-41.153
-49.052

-4.365
-9.592
-12.704
-20.188
-25.378
-29.242
-37.050

-3.266
-7.418
-10.044
-15.682
-18.636
-22.326
-28.839

1
5
10
21
31
42
63

53

Table 2.4: Relative Abnormal Performance of Local and Distant Mutual Funds Following Stress
Events

The following tables report estimated cumulative abnormal returns and simulated returns for U.S.
based equity mutual fund managers following stressful events. Panel A reports results using a
sample of 361 U.S. based foreign equity mutual funds located within 100 miles of stressful events
bench-marked against a matched sample of funds headquartered 1,000 miles from the event. Panel
B reports results for 1,081 domestic equity mutual fund managers located with 100 miles of the
traumatic events. Cumulative abnormal returns for the ‘near’ and ‘distant’ group of managers were
calculated using the Fama French 3 Factors calculated over the quarter preceding each traumatic
event. Abnormal returns for both groups are benchmarked against a matched sample of distant
mutual funds. The matching criteria used is: CRSP 3 digit objective code, Monthly Total Net
Assets, Quarterly Returns, Fund Turnover, and Fund Age. For each event day within the event
window, the abnormal return is calculated as the mean abnormal return across each of the 20 events
in the sample. Cumulative Abnormal Return (CAR) is calculated as the cumulative product of event
days from t0 to time t and reported in basis points. Columns (3) to (6) report simulated cumulative
abnormal returns based on 1,500 random draws with randomized event days and randomized event
order taken to match the original distribution of events across the sample period.

Panel A: International Equity Funds (Near less Far bps)

Event_Time (days) Actual (bps) Simulated_03 Simulated_05 Simulated_10 Simulated_50
0.015
0.534
0.405
1.534
3.297
4.493
8.374

-11.021
-24.760
-35.129
-47.058
-58.645
-69.978
-91.805

-9.448
-21.182
-30.298
-42.269
-52.155
-61.069
-72.628

-7.222
-15.929
-23.336
-32.470
-40.572
-44.833
-55.835

0.365
13.930
15.856
-0.724
-47.030
-70.204
-80.936

1
5
10
21
31
42
63

Panel B: Domestic Equity Funds (Near less Far bps)

Event_Time (days) Actual (bps) Simulated_03 Simulated_05 Simulated_10 Simulated_50
0.053
-0.075
-0.016
-0.411
-0.339
-0.965
-1.500

-4.818
-10.988
-14.920
-24.342
-28.905
-36.143
-41.938

-6.069
-11.880
-20.181
-23.608
-32.478
-61.497
-69.380

-4.365
-9.592
-12.704
-20.188
-25.378
-29.242
-37.050

-3.266
-7.418
-10.044
-15.682
-18.636
-22.326
-28.839

1
5
10
21
31
42
63

54

Table 2.5: Mutual Funds Change in Risk Exposure Following Traumatic Events

The following tables report regressions on the change in 3 Factor betas before and after traumatic
events. Fama French 3 Factor coeﬃcients are calculated over the quarter preceding each traumatic
event and post-event factors use the quarter following each event. The three columns report the
Market beta coeﬃcients, SMB coeﬃcients, and HML coeﬃcients respectively. The regressions are
estimated as a Diﬀerence-in-Diﬀerence Model between ‘near’ funds located 100 miles from the
traumatic event and ‘distant’ funds located 1,000 miles from the event. The variable of interest is
the interaction between Post Event and Near funds which captures whether funds located in close
proximity to traumatic events change their risk exposure to a greater extent than distant funds.
Control variables included are fund size, fund age, 12 month turnover and exposure to the other
two risk factors. Event eﬀects and 3 digit CRSP objective codes are included in each speciﬁcation.
Robust standard errors are reported in parentheses below the coeﬃcient estimates. Statistical
signiﬁcance (two-sided) at the 10%, 5%, and 1% levels is denoted by *, **, and ***, respectively.
Panel (A) reports results for Domestic Equity funds and Panel (B) reports results for International
Equity funds. Panel (C) reports simulated results for changes in market betas using 1,500 random
draws for event dates and locations.

Panel A: Domestic Equity Funds

Post_Event

Near

Post_Event x Near

Turnover

Pre_SMB

Pre_HML

Pre_Mom

Pre_MKT

Mkt Beta
−0.2155∗
(−1.7250)
0.0001
(0.5231)
−0.0942
(−0.6067)
0.0032
(1.1701)
0.0307
(0.2230)
−0.2746∗
(−1.8264)
−0.3702∗
(−1.8271)

Event Eﬀects
Investment Style Eﬀects
R2
Adj. R2
Num. obs.
RMSE
∗∗∗p < 0.01, ∗∗p < 0.05, ∗p < 0.1

Yes
Yes
0.1373
0.1295
3215
0.0059

SMB Beta HML Beta Mom Beta
0.0995
0.0499
(0.4232)
(0.6667)
0.0001
0.0001
(0.5220)
(0.4723)
−0.0850
−0.1207
(−1.2919)
(−0.3127)
0.0014
0.0032
(0.5848)
(1.4138)
−0.0063
(−0.0630)
−0.1503
(−1.1174)

−0.0317
(−0.7285)
0.0000
(0.3677)
−0.0590
(−0.9777)
0.0008
(0.4306)
−0.0033
(−0.0990)

−0.0979
(−1.1899)
−0.0854∗
(−1.8877)

Yes
Yes
0.1664
0.1588
3215
0.0033

−0.1838∗∗
(−2.4437)

Yes
Yes
0.1196
0.1115
3215
0.0042

0.0115
(0.1454)
−0.0259
(−0.2147)
0.0360
(0.3589)

Yes
Yes
0.1182
0.1102
3215
0.0052

55

Table 2.5 (cont’d): Mutual Funds Change in Risk Exposure Following Traumatic Events

Panel B: International Equity Funds

SMB Beta HML Beta Mom Beta
0.3896∗∗∗
0.1314
(3.9105)
(1.0835)
−0.0005
0.0005
(−1.2067)
(0.1074)
−0.2268
−0.1460
(−1.0711)
(−1.6202)
−0.0012
−0.0115
(−0.0340)
(−0.3665)
−0.0034
(−0.0222)
−0.1619∗
(−1.5760)

0.0298
(0.5518)
−0.0004
(−1.4099)
0.0326
(0.6898)
−0.0276
(−1.0403)
−0.0235
(−0.3614)

Post_Event

Near

Post_Event x Near

Turnover

Pre_SMB

Pre_HML

Pre_Mom

Pre_MKT

Mkt Beta
−0.0599
(−0.9975)
0.0002
(0.7211)
−0.0678
(−0.9929)
−0.0229
(−0.8719)
−0.1449∗∗
(−2.5113)
−0.3115∗∗∗
(−2.6108)
−0.1325∗∗
(−2.1208)

Event Eﬀects
Investment Style Eﬀects
R2
Adj. R2
Num. obs.
RMSE
∗∗∗p < 0.01, ∗∗p < 0.05, ∗p < 0.1

Yes
Yes
0.4823
0.4709
1115
0.0058

−0.0337
(−0.2502)
−0.0266
(−0.2532)
−0.1980∗
(−1.7278)

Yes
Yes
0.2922
0.2766
1115
0.0067

−0.0913∗∗
(−2.3654)
−0.1999∗∗∗
(−4.2988)

Yes
Yes
0.3154
0.3003
1115
0.0045

−0.1936∗
(−1.8537)

Yes
Yes
0.5346
0.5244
1115
0.0066

Panel C: Actual and Simulated Change in Market Beta Following Traumatic Events

Near International
Far International
Near Domestic
Far Domestic

Actual
0.003
0.003
-0.003
-0.01

3%
-0.372
-0.371
-0.268
-0.250

5%
-0.320
-0.316
-0.208
-0.198

10%
-0.249
-0.243
-0.134
-0.131

56

50% 90% 95% 97%
0.389
0.383
0.299
0.278

0.240
0.242
0.132
0.129

0.331
0.328
0.217
0.205

-0.006
-0.004
-0.002
-0.002

Table 2.6: Mutual Fund Change in Turnover Following Traumatic Events

The following tables report the change in fund asset turnover following traumatic events. Fund
Turnover is taken from the CRSP database and reported on a quarterly basis.
It is calculated
within CRSP as the prior 12 month asset purchases divided by total net assets. For each event I
identify the most recently reported turnover ﬁgure before and after each event and test for changes
in manager purchasing activity in the months following traumatic events. I use a Diﬀerence-in-
Diﬀerences approach using ‘near’ funds located 100 miles from the traumatic event and ‘distant’
funds located 1,000 miles from the event. The variable of interest is the interaction between Post
Event and Near funds which captures whether funds located in close proximity to traumatic events
change their purchasing activity following traumatic events. The sample of mutual fund managers
is split between high and low turnover funds. Funds in the bottom quartile of turnover within the
sample are considered low-turnover funds, and funds in the highest quartile are high-turnover funds.
Columns (1) and (2) report results on the high and low turnover subsamples. Column (3) reports
the combined results using an indicator for the low-turnover group. Control variables included
are fund size and fund age. Event eﬀects and 3 digit CRSP objective codes are included in each
speciﬁcation. Robust standard errors are reported in parentheses below the coeﬃcient estimates.
Statistical signiﬁcance (two-sided) at the 10%, 5%, and 1% levels is denoted by *, **, and ***,
respectively. Panel (A) reports results for Domestic Equity funds and Panel (B) reports results for
International Equity funds. Panel (C) reports simulated results for changes in turnover using 1,500
random draws for event dates and locations.

Panel A: Domestic Mutual Fund Turnover

Post_Event

Near

Post_Event x Near

log(Fund Size)

Fund Age

1.1083∗∗∗
(4.8824)
2.5564∗∗∗
(3.3440)
−0.8880∗∗∗
(−3.6766)
0.1597∗∗
(1.9809)
0.0234
(0.8703)

Low Turnover High Turnover Diﬀ ( χ2)
0.5070∗∗∗
−0.601
(0.160)
(5.5932)
0.0574∗∗
−2.499∗∗∗
(66.061)
(2.5977)
−0.4256∗∗∗
0.462
(−4.0650)
(.267)
−0.161∗∗∗
−0.0010∗∗
(−2.1972)
(8.843)
−0.022
0.0017
(2.362)
(1.3544)
Yes
Yes
0.7316
0.2020
1199

Yes
Yes
0.8159
0.3803
617

Event Eﬀects
Investment Style Eﬀects
R2
Adj. R2
Num. obs.
∗∗∗p < 0.01, ∗∗p < 0.05, ∗p < 0.1

Yes
Yes
0.3142
-1.0752

582

57

Table 2.6 (cont’d): Mutual Fund Change in Turnover Following Traumatic Events

Panel B: International Mutual Fund Turnover

Post_Event

Near

Post_Event x Near

log(Fund Size)

Fund Age

Event Eﬀects
Investment Style Eﬀects
R2
Adj. R2
Num. obs.
∗∗∗p < 0.01, ∗∗p < 0.05, ∗p < 0.1

1.0507∗∗∗
(10.9762)
0.3389∗∗
(2.4514)
−0.2147∗
(−1.8811)
0.0100
(0.0948)
0.0072
(0.6461)

Low Turnover High Turnover Diﬀ ( χ2)
0.7616∗∗∗
−0.289∗
(2.619)
(10.5929)
−0.300
0.0390
(0.302)
(1.4393)
−0.055∗∗
−0.2700
(−1.3607)
(3.549)
−0.010
0.0003
(.368)
(1.3253)
−0.051
0.0006
(0.2376)
(1.170)
Yes
Yes
0.8543
0.5586
404

Yes
Yes
0.8458
0.4168
209

Yes
Yes
0.6745
-0.0185

195

Panel C: Combined Simulated Sample

10% 50% 90% 95% 97%
0.289
0.267
0.340
0.320

0.087
0.080
0.100
0.090

0.190
0.170
0.220
0.202

0.000
0.000
0.000
0.000

-0.067
-0.068
-0.080
-0.073

Near International
Far International
Near Domestic
Far Domestic

Actual
0.194
-0.002
0.407
0.0458

3%
-0.240
-0.240
-0.300
-0.287

5%
-0.150
-0.150
-0.190
-0.170

58

Table 2.7: Mutual Fund Tracking Errors

The following tables report results on mutual fund benchmark tracking errors. Fund benchmarks
are identiﬁed as the index with the highest R2 to the fund when regression daily returns onto a set of
indices from the St. Louis Federal Reserve. Tracking errors are calculated as the standard deviation
of the diﬀerence in returns between the mutual fund and its benchmark index. Tracking errors
are calculated the period t + 21 to t + 63 and t − 21 to t − 63 to control for any contemporaneous
eﬀects from the initial event itself. Funds are split along the dimension of pre-event tracking
errors. Funds in the lowest quartile are considered ‘closet-index’ funds(Column 1) and funds in
the highest quartile are considered ‘Active’ funds(Column 2). I use a Diﬀerence-in-Diﬀerences
approach using ‘near’ funds located 100 miles from the traumatic event and ‘distant’ funds located
1,000 miles from the event. The variable of interest is the interaction between Post Event and Near
funds which captures whether funds located in close proximity to traumatic events change their
indexing strategy following traumatic events. Column (3) reports the combined sample with a triple
interaction to capture diﬀerences in behavior between ‘closet-index’ and ‘active’ funds. Control
variables included are fund size and fund turnover. Event eﬀects and 3 digit CRSP objective codes
are included in each speciﬁcation. Robust standard errors are reported in parentheses below the
coeﬃcient estimates. Statistical signiﬁcance (two-sided) at the 10%, 5%, and 1% levels is denoted
by *, **, and ***, respectively.

Post_Event

Near

Post_Event x Near

log(Fund Size)

Turnover

Event Eﬀects
Investment Style Eﬀects
R2
Adj. R2
Num. obs.
∗∗∗p < 0.01, ∗∗p < 0.05, ∗p < 0.1

Yes
Yes
0.7309
0.7242
662

Indexing
0.6226∗∗∗
(18.1494)
0.0134∗∗
(2.2165)
−0.0803∗
(−1.7418)
−0.0007
(−1.1665)
0.0023∗∗
(2.1298)

Active
0.6566∗∗∗
(22.1085)
−0.1104∗∗∗
(−2.6051)
0.1761∗∗
(2.5464)
−0.0069
(−1.4769)
0.0120
(1.5072)

Yes
Yes
0.8689
0.8650
621

Diﬀ ( χ2)
−0.034
(0.052)
0.124∗∗∗
(20.332)
−0.256∗
(2.698)
0.006
(1.699)
−0.010∗∗
(3.311)
Yes
Yes
0.9198
0.9185
1283

59

APPENDIX C

MUTUAL FUND CLIENTS

60

Figure 3.1: Flow-Performance Sensitivity: Normal vs Stress

This ﬁgure reports the ﬂow performance sensitivity for mutual funds separately for the normal and
high stress periods. Flow performance sensitivity is estimated for each past performance decile
using past year returns. The grey line depicts fund ﬂows for the normal periods and the dashed
orange line reports the high stress periods.

61

2342.55.07.510.0Return Rankings: Low to HighFund Flow(%)NormalStressMutual Fund Flow − Performance Sensitivty with StressTable 3.1: Summary Stats

The following reports summary statistics for my two primary sample data sets. Panel A reports a
constructed sample of mutual funds from the CRSP universe aggregated to the fund series level
using a mapping of CRSP fundnos to EDGAR Series CIKs. This sample contains 7,928 unique
fund series from 1,030 diﬀerent fund companies. Panel B reports a second sample of funds
dis-aggregated at the share-calss level which covers the same set of CRSP fund series in Panel A.

Panel A: Fund Series

Flow_t
Monthly Return
Monthly Total Net Assets
Expense Ratio
12b1 Fees
Management Fee
Fund Age (Months)

Mean Median Std. Deviation
0.169
0.020
0.000
0.000
1683.846
6987.329
0.011
0.008
0.002
0.001
0.007
0.005
157.845
131.804

0.007
0.000
309.800
0.007
0.000
0.004
126.000

P10
-0.059
-0.000
33.631
0.002
0.000
0.001
43.000

P90
0.087
0.000
3177.000
0.014
0.002
0.009
292.000

Panel B: Share Class Level Data

Flow_t
Monthly Return
Monthly Total Net Assets
Expense Ratio
12b1 Fees
Management Fee
Fund Age (Months)

Mean Median Std. Deviation
0.155
0.007
0.041
0.005
761.233
3258.440
0.006
0.011
0.004
0.003
0.003
0.006
131.964
102.521

-0.003
0.004
123.900
0.010
0.000
0.006
107.000

P10
-0.040
-0.035
18.700
0.003
0.000
0.001
40.000

P90
0.050
0.049
1387.000
0.019
0.010
0.010
244.000

62

Table 3.2: Mutual Fund Flow Sensitivity to Search Costs During Normal and Stress Periods

The following table reports the OLS estimated coeﬃcients and t-statistics for fund ﬂows for separate
regressions between normal and high stress periods following nationally reported traumatic events.
The variables of interest here are log(Monthl yTotalNet Assets) and 12b1Fees which are proxies
for mutual fund search costs. Panel (A) reports results using past year returns to assign funds into
quintiles. Panel (B) repeats the same tests using past quarter returns. Column (1) displays results
for normal periods. Column(2) shows results using only months identiﬁed as high stress periods.
Column (3) reports the diﬀerence in coeﬃcient estimates and the χ2 test under the null of no
diﬀerence between the group coeﬃcients. Year eﬀects and 3 digit CRSP objective code eﬀects are
included in each speciﬁcation. Standard errors are reported in parentheses below the coeﬃcient
estimate and are clustered at the fund-series level in every speciﬁcation. Statistical signiﬁcance
(two-sided) at the 10%, 5%, and 1% levels is denoted by *, **, and ***, respectively.

Panel A: Past Year Returns

(1)

(2)

Past Year Ret. Rankings Past Year Ret. Rankings

(Intercept)

Return Quintile 1

Return Quintile 2

Return Quintile 4

Return Quintile 5

log(Monthly Total Net Assets)

log(Fund Age)

Return Risk

Actual 12b1 fees

Management Fee

log(fund family size)

Return Quintile 1 - Return Quintile 5

Year Eﬀects
Investment Style Eﬀects
Fund Series Clustered Errors
Adj. R2
Num. obs.
∗∗∗p < 0.01, ∗∗p < 0.05, ∗p < 0.1

Normal Period
0.0500∗∗∗
(4.3229)
−0.0051∗∗∗
(−7.6603)
−0.0022∗∗∗
(−4.9125)
−0.0002
(−0.4497)
0.0225∗∗∗
(26.0246)
−0.0093∗∗∗
(−27.7274)
−0.0026∗∗∗
(−5.7654)
13.3324∗∗∗
(6.3821)
−1.5599∗∗∗
(−7.2158)
−0.2374∗
(−1.6994)
0.0034∗∗∗
(17.0505)
−0.0276∗∗∗
(695.76)

Yes
Yes
Yes
0.0213
851948

63

Stress Events
0.0444∗∗∗
(3.9314)
−0.0087∗∗∗
(−7.4135)
−0.0038∗∗∗
(−4.2411)
0.0030∗∗∗
(3.1598)
0.0131∗∗∗
(9.9391)
−0.0063∗∗∗
(−15.0288)
−0.0032∗∗∗
(−5.2426)
6.3487∗∗∗
(2.6901)
0.6949∗∗∗
(2.7861)
−0.3727∗∗∗
(−3.6544)
0.0019∗∗∗
(8.2693)
−0.0218∗∗∗
(208.18)

Yes
Yes
Yes
0.0141
211225

(3)
Diﬀ
χ2

0.004∗∗∗
(7.489)
0.002
(1.468)
−0.003∗∗
(6.029)
0.009∗∗∗
(49.461)
−0.003∗∗∗
(116.771)
0.002
(1.012)
6.032∗∗∗
(39.348)
−2.175∗∗∗
(82.555)
0.110
(2.119)
0.001∗∗∗
(35.165)
−0.005∗∗∗
(19.343)

Table 3.2 (cont’d): Mutual Fund Flow Sensitivity to Search Costs During Normal and Stress Periods

Panel B: Past Quarter Returns

(3)
Diﬀ
χ2

−0.003∗∗
(4.457)
−0.002
(2.372)
0.000
(.111)
0.007∗∗∗
(26.607)
−0.004∗∗∗
(169.651)
0.001
(1.029)
8.979∗∗∗
(65.089)
−2.376∗∗∗
(91.579)
0.096
(1.075)
0.001∗∗∗
(33.760)
−0.004∗∗∗
(53.488)

(1)

(2)

Past Qtr Ret. Rankings Past Qtr Ret. Rankings

(Intercept)

Return Quintile 1

Return Quintile 2

Return Quintile 4

Return Quintile 5

log(Monthly Total Net Assets)

log(Fund Age)

Return Risk

Actual 12b1 fees

Management Fee

log(fund family size)

Return Quintile 1 - Return Quintile 5

Year Eﬀects
Investment Style Eﬀects
Fund Series Clustered Errors
Adj. R2
Num. obs.
∗∗∗p < 0.01, ∗∗p < 0.05, ∗p < 0.1

Stress Events
0.0424∗∗∗
(3.7239)
0.0006
(0.4391)
0.0005
(0.4424)
0.0009
(0.8205)
0.0059∗∗∗
(4.2655)
−0.0061∗∗∗
(−14.7083)
−0.0042∗∗∗
(−6.7898)
7.3001∗∗∗
(2.8333)
0.6153∗∗
(2.4308)
−0.3473∗∗∗
(−3.3520)
0.0020∗∗∗
(8.3499)
.0001∗∗∗
(18.194)

Yes
Yes
Yes
0.0123
211225

Normal Period
0.0507∗∗∗
(4.3394)
0.0014∗∗
(1.9715)
0.0033∗∗∗
(4.8921)
0.0041∗∗∗
(6.1405)
0.0155∗∗∗
(18.0686)
−0.0097∗∗∗
(−28.3482)
−0.0036∗∗∗
(−7.7775)
16.2793∗∗∗
(5.9781)
−1.7611∗∗∗
(−7.7508)
−0.2509∗
(−1.7101)
0.0034∗∗∗
(16.7654)
−0.0141∗∗∗
(326.47)

Yes
Yes
Yes
0.0191
851948

64

Table 3.3: Mutual Fund Flow Sensitivity to Past Performance During Normal and Stress Periods

The following table reports the OLS estimated coeﬃcients and t-statistics for fund ﬂows for separate
regressions between normal and high stress periods following nationally reported traumatic events.
This table estimates diﬀerences in fund ﬂow - performance sensitivity between normal and high
stress periods. Panel (A) reports results using past year returns to assign funds into quintiles.
Panel (B) repeats the same tests using past quarter returns. Column (1) displays results for normal
periods. Column(2) shows results using only months identiﬁed as high stress periods. Column (3)
reports the diﬀerence in coeﬃcient estimates and the χ2 test under the null of no diﬀerence between
the group coeﬃcients. Return quintile 1 groups the lowest performing funds in each period and
Quintile 5 groups the highest performing funds. Funds are ranked and assigned quintiles separately
for each CRSP fund obejctive identiﬁer. Year eﬀects and 3 digit CRSP objective code eﬀects are
included in each speciﬁcation. Standard errors are reported in parentheses below the coeﬃcient
estimate and are clustered at the fund-series level in every speciﬁcation. Statistical signiﬁcance
(two-sided) at the 10%, 5%, and 1% levels is denoted by *, **, and ***, respectively.

Panel A: Past Year Returns

(1)

(2)

Past Year Ret. Rankings Past Year Ret. Rankings

Return Quintile 1

Return Quintile 2

Return Quintile 4

Return Quintile 5

Expense Ratio

log(Monthly Total Net Assets)

log(Fund Age)

Return Risk

log(fund family size)

Return Quintile 1 - Return Quintile 5

Year Eﬀects
Investment Style Eﬀects
Adj. R2
Num. obs.
∗∗∗p < 0.01, ∗∗p < 0.05, ∗p < 0.1

Normal Period
−0.0051∗∗∗
(−7.7722)
−0.0023∗∗∗
(−4.9861)
−0.0002
(−0.4065)
0.0226∗∗∗
(26.0650)
−0.2268∗∗
(−2.0524)
−0.0093∗∗∗
(−27.6638)
−0.0026∗∗∗
(−5.7607)
13.4865∗∗∗
(6.3691)
0.0035∗∗∗
(17.0969)
−0.0277∗∗∗
(700.55)

Yes
Yes
0.0211
852030

65

Stress Events
−0.0088∗∗∗
(−7.4884)
−0.0038∗∗∗
(−4.2279)
0.0029∗∗∗
(3.1117)
0.0130∗∗∗
(9.8112)
−0.0320∗∗∗
(−3.2446)
−0.0063∗∗∗
(−15.0158)
−0.0032∗∗∗
(−5.2125)
6.1638∗∗∗
(2.6450)
0.0020∗∗∗
(8.6053)
−0.0218∗∗∗
(207.3)
Yes
Yes
0.0140
211237

(3)
Diﬀ
χ2
0.004∗∗∗
(7.586)
0.002
(1.379)
−0.003∗∗
(5.783)
0.010∗∗∗
(52.202)
−0.003∗∗∗
(41.330)
0.004∗∗∗
(122.212)
0.002
(0.838)
7.3337∗∗∗
4(3.338)
0.010∗∗∗
(36.963)
−0.006∗∗∗
(20.914)

Table 3.3 (cont’d): Mutual Fund Flow Sensitivity to Past Performance During Normal and Stress
Periods

Panel B: Past Quarter Returns

(3)
Diﬀ
χ2
−0.003∗∗
(4.198)
−0.002
(2.437)
0.000
(0.139)
0.007∗∗∗
(27.978)
0.000∗∗∗
(53.162)
−0.003∗∗∗
(176.947)
−0.002
(0.835)
9.3525∗∗∗
(70.752)
0.007∗∗∗
(36.105)
−0.01∗∗∗
(54.487)

(1)

(2)

Past Qtr Ret. Rankings Past Qtr Ret. Rankings

Return Quintile 1

Return Quintile 2

Return Quintile 4

Return Quintile 5

Expense Ratio

log(Monthly Total Net Assets)

log(Fund Age)

Return Risk

log(fund family size)

Return Quintile 1 - Return Quintile 5

Year Eﬀects
Investment Style Eﬀects
Fund Series Clustered Errors
Adj. R2
Num. obs.
∗∗∗p < 0.01, ∗∗p < 0.05, ∗p < 0.1

Stress Events
−0.0006
(−0.5216)
0.0000
(0.0476)
0.0004
(0.4373)
0.0053∗∗∗
(4.2693)
−0.0348∗∗∗
(−3.0048)
−0.0061∗∗∗
(−14.6989)
−0.0042∗∗∗
(−6.7699)
7.1280∗∗∗
(2.7999)
0.0020∗∗∗
(8.6633)
−0.0059∗∗∗
(18.273)

Yes
Yes
Yes
0.0140
211237

Normal Period
−0.0033∗∗∗
(−4.9096)
−0.0020∗∗∗
(−4.0491)
0.0009∗
(1.7902)
0.0123∗∗∗
(16.2215)
−0.2559∗∗
(−2.0627)
−0.0097∗∗∗
(−28.2118)
−0.0036∗∗∗
(−7.7323)
16.4805∗∗∗
(5.9641)
0.0035∗∗∗
(16.6361)
−0.0156∗∗∗
(329.33)

Yes
Yes
Yes
0.0211
852030

66

Table 3.4: Retail Share-Class Sensitivity to Past Performance During Normal and Stress Periods

The following table reports the OLS estimated coeﬃcients and t-statistics for retail mutual fund
share-class fund ﬂows for separate regressions between normal and high stress periods following
nationally reported traumatic events. This table estimates diﬀerences in fund ﬂow - performance
sensitivity between normal and high stress periods. Panel (A) reports results using past year returns
to assign funds into quintiles. Panel (B) repeats the same tests using past quarter returns. Column
(1) displays results for normal periods. Column(2) shows results using only months identiﬁed
as high stress periods. Column (3) reports the diﬀerence in coeﬃcient estimates and the χ2 test
under the null of no diﬀerence between the group coeﬃcients. The variables of interest here are
the reported past return quintile indicators which estimate the convexity of the ﬂow-performance
sensitivity. Return Quintile 3 is used as a baseline for the return ranking indicators. Return quintile
1 groups the lowest performing funds in each period and Quintile 5 groups the highest performing
funds. Funds are ranked and assigned quintiles separately for each CRSP fund objective identiﬁer.
Year eﬀects and 3 digit CRSP objective code eﬀects are included in each speciﬁcation. Standard
errors are reported in parentheses below the coeﬃcient estimate and are clustered at the fund-series
level in every speciﬁcation. Statistical signiﬁcance (two-sided) at the 10%, 5%, and 1% levels is
denoted by *, **, and ***, respectively.

Panel A: Retail Past Quarter Returns

(1)

(2)

Past Qtr Ret. Rankings Past Qtr Ret. Rankings

Return Quintile 1

Return Quintile 2

Return Quintile 4

Return Quintile 5

Expense Ratio

log(Monthly Total Net Assets)

log(Fund Age)

Return Risk

log(fund family size)

Return Quintile 1 - Return Quintile 5

Year Eﬀects
Investment Style Eﬀects
Adj. R2
Num. obs.
∗∗∗p < 0.01, ∗∗p < 0.05, ∗p < 0.1

Normal Period
−0.0041∗∗∗
(−8.9057)
−0.0025∗∗∗
(−7.0551)
0.0034∗∗∗
(9.4580)
0.0166∗∗∗
(31.2768)
−0.9383∗∗∗
(−18.4022)
−0.0031∗∗∗
(−15.2625)
−0.0102∗∗∗
(−32.6384)
0.0739∗∗
(1.9732)
−0.0003∗∗
(−2.3666)
−0.0207∗∗∗
(2002)
Yes
Yes
0.0105
1302261

67

Stress Events
−0.0019∗
(−1.9444)
−0.0028∗∗∗
(−3.4563)
−0.0000
(−0.0004)
0.0091∗∗∗
(9.7421)
−0.8245∗∗∗
(−9.5077)
−0.0032∗∗∗
(−9.7028)
−0.0083∗∗∗
(−15.6979)
0.3228∗∗∗
(3.7725)
−0.0002
(−1.0483)
−0.011∗∗∗
(426.16)

Yes
Yes
0.0075
258288

(3)
Diﬀ
χ2
−0.002∗∗
(5.442)
0.000
(0.097)
0.003∗∗∗
(12.691)
0.007∗∗∗
(59.535)
0.003∗
(2.964)
−0.002
(0.223)
0.000∗∗∗
(14.521)
−.2489
(82.319)
0.007
(0.269)
−0.009∗
(3.6997)

Table 3.4 (cont’d): Retail Share-Class Sensitivity to Past Performance During Normal and Stress
Periods

Panel B: Institutions Past Quarter Returns

(3)
Diﬀ
χ2
0.002∗
(2.740)
−0.002∗
(2.970)
0.000
(0.025)
0.001
(1.093)
0.000
(0.025)
0.002∗∗∗
(31.347)
−0.002
(1.324)
(10.42)
0.001
(1.093)
0.001∗∗∗
(102.78)

(1)

(2)

Past Qtr Ret. Rankings Past Qtr Ret. Rankings

Return Quintile 1

Return Quintile 2

Return Quintile 4

Return Quintile 5

Expense Ratio

log(Monthly Total Net Assets)

log(Fund Age)

Return Risk

log(fund family size)

Return Quintile 1 - Return Quintile 5

Year Eﬀects
Investment Style Eﬀects
Adj. R2
Num. obs.
∗∗∗p < 0.01, ∗∗p < 0.05, ∗p < 0.1

Stress Events
−0.0071∗∗∗
(−6.0590)
−0.0003
(−0.2307)
0.0030∗∗∗
(2.6326)
0.0117∗∗∗
(9.5196)
−1.2289∗∗∗
(−10.3924)
−0.0045∗∗∗
(−14.4932)
−0.0131∗∗∗
(−17.9946)
0.1615∗∗∗
(3.8845)
0.0009∗∗∗
(3.5824)
−0.0188∗∗∗
(112.86)

Yes
Yes
0.0080
195646

Normal Period
−0.0048∗∗∗
(−6.6678)
−0.0027∗∗∗
(−4.5505)
0.0032∗∗∗
(5.2088)
0.0132∗∗∗
(17.9568)
−1.2204∗∗∗
(−14.8117)
−0.0062∗∗∗
(−23.8693)
−0.0139∗∗∗
(−29.6529)
0.0668∗
(1.8890)
0.0013∗∗∗
(6.9611)
−0.018∗∗∗
(1334.5)

Yes
Yes
0.0095
754396

68

Table 3.5: Smart Money: 24 month Fund Flow Weighted Alphas after Traumatic Events

In each month I construct high and low ﬂow portfolios based on fund ﬂows to retail and institutional
share classes for actively managed equity mutual funds. I identify if the current periods fund ﬂows
are abnormally high or low(greater/less than 1.5 standard deviations above the rolling 36 month
mean) and from this construct value weighted portfolios separately for retail and institutional share-
classes which are then held for 24 months. I compute the 4 factor alpha for the portfolio constructed
in each month to generate a time series of alphas representing the aggregate decisions made by
investors. In Panel A, I report portfolio alpha means, diﬀerences, and t-tests between normal and
stress periods. T-statistics are reported in parentheses. Panel B reports simulated diﬀerences in
alphas between normal and randomized event days using a series of 3,000 simulated draws. The
ﬁrst 3 rows report mean normal period and stress period alphas along with the diﬀerence in alphas
for retail and institutional share classes. The next group of rows report the simulated distribution
for diﬀerences in normal and randomized stress event alphas.

Panel A: Summary Tests for Normal and Stress Period Alphas

Retail

Institutional

Diﬀ

Normal Alpha Stress Alpha
.0012
(3.1830)
0.0008
(2.471)
.0004
(0.70083)

0.0002
(0.2493)
-0.0009
(−1.204)
.0011
(1.0951)

Diﬀ
.0010
(1.500)
.0017
(2.132)

Panel B: Simulated Diﬀerences between Normal and Stress Portfolio Alpha

Retail
Institutional

Actual_Diﬀ Simulated_97 Simulated_95 Simulated_90 Simulated_50 Simulated_10
−0.0006
−0.0006

0.0010
0.0017∗∗

0.0005
0.0006

0.0019
0.0019

0.0017
0.0017

0.0015
0.0015

69

Table 3.6: NSAR: Stress Eﬀects on Mutual Fund Inﬂow and Outﬂows

The following table shows the breakdown of net mutual fund ﬂows into their inﬂow (Column 1)
and outﬂow (Column 2) components. The data for this table is taken from the NSAR ﬁlings from
EDGAR with date back to 2006. Only funds whose NSAR ﬂows have a correlation of .9 with
CRSP are included in the sample used to construct the following table. This sample includes 2,320
unique fund series. The model regresses ﬂows in month t onto an indicator to identify if month t
is a high stress month or not. All other variables are lagged including past net ﬂows. Year eﬀects
and 3 digit CRSP objective code eﬀects are included in each speciﬁcation. Standard errors are
reported in parentheses below the coeﬃcient estimate and are clustered at the fund-series level in
every speciﬁcation. Statistical signiﬁcance (two-sided) at the 10%, 5%, and 1% levels is denoted
by *, **, and ***, respectively.

Stress_Ind

Return Quintile 1

Return Quintile 2

Return Quintile 4

Return Quintile 5

Expense Ratio

log(Monthly Total Net Assets)

log(Fund Age)

Return Risk

log(Fund Family Size)

Year Eﬀects
Investment Style Eﬀects
Fund Share Class Clustered Errors
R2
Adj. R2
Num. obs.
∗∗∗p < 0.01, ∗∗p < 0.05, ∗p < 0.1

(2)

(3)

(1)

Net ﬂow
Inﬂow
−0.0021∗∗∗
−0.0030∗∗∗
(−4.7816)
(−3.9654)
−0.0068∗∗∗
−0.0059∗∗∗
(−7.4745)
(−2.6051)
−0.0018∗∗∗
−0.0024
(−2.6937)
(−1.4613)
0.0036∗∗
0.0010
(1.3781)
(2.0163)
0.0027∗∗∗
0.0022
(2.7920)
(0.9627)
−0.2320∗∗
−0.6476∗
(−1.9768)
(−1.7711)
−0.0226∗∗∗ −0.0181∗∗∗ −0.0028∗∗∗
(−6.3678)
(−16.2701)
−0.0056∗∗∗
−0.0019
(−1.0553)
(−8.7968)
−6.3713∗∗
−15.2956
(−2.4202)
(−1.6099)
0.0074∗∗∗
0.0007∗
(7.8074)
(1.9435)

Outﬂow
0.0006
(1.1743)
0.0018
(1.0256)
−0.0007
(−0.6097)
0.0030∗∗
(2.1254)
−0.0003
(−0.1457)
−0.5288∗∗
(−2.0851)
(−17.4269)
0.0046∗∗∗
(3.0877)
−10.0676
(−1.3713)
0.0062∗∗∗
(8.1288)

Yes
Yes
Yes
0.1021
0.1016
91702

Yes
Yes
Yes
0.1158
0.1153
91817

Yes
Yes
Yes
0.0244
0.0239
89902

70

Table 3.7: Relative Daily Performance of Local and Distant Mutual Funds Following Stress Events

The following tables report estimated cumulative abnormal returns and simulated returns for U.S.
based equity mutual fund managers following stressful events. Panel A reports results using a
sample of 743 U.S. based foreign equity mutual funds located within 100 miles of stressful events
bench-marked against a matched sample of funds headquartered 1,000 miles from the event. Panel
B reports results for 1,825 domestic equity mutual fund managers located with 100 miles of the
traumatic events. Daily returns for both groups are benchmarked against a matched sample of
distant mutual funds. The matching criteria used is: CRSP 3 digit objective code, Monthly Total
Net Assets, Quarterly Returns, Fund Turnover, and Fund Age. For each event day within the event
window, the abnormal return is calculated as the mean abnormal return across each of the 20 events
in the sample. Cumulative Abnormal Return (CAR) is calculated as the cumulative product of event
days from t0 to time t and reported in basis points. Columns (3) to (5) report simulated cumulative
abnormal returns based on 3,000 random draws with randomized event days and randomized event
order taken to match the original distribution of events across the sample period.

Panel A: Domestic Equity Funds

Event_Time CAR(Near less Far) Simulated_03 Simulated_05 Simulated_10 Simulated_50
-0.1
-0.8
-1.5
-1.5
-3.2
-3.9
-3.7
-4.1
-3.9
-4.4
-4.5
-4.5
-4.4
-4.1

-9.3
-21.9
-31.1
-39.2
-43.7
-51.3
-55.3
-60.3
-65.7
-67.8
-72.3
-76.4
-78.9
-81.1

-7.2
-16.9
-24.7
-30.3
-35.2
-39.7
-44.1
-47.4
-51.5
-54.5
-58.1
-60.5
-62.0
-63.2

1
7
14
21
28
35
42
49
56
63
70
77
84
91

4.6
10.1
-22.0
-19.0
-24.0
-33.2
-55.1
-52.0
-68.4
-74.0
-78.6
-80.0
-93.7
-90.8

-11.1
-25.6
-35.6
-45.0
-52.1
-58.0
-64.5
-68.9
-74.1
-78.1
-81.9
-87.3
-90.7
-92.5

71

Table 3.7 (cont’d): Relative Daily Performance of Local and Distant Mutual Funds Following Stress
Events

Panel B: Foreign Equity Funds

Event_Time CAR(Near less Far) Simulated_03 Simulated_05 Simulated_10 Simulated_50
-0.1
-0.6
-0.3
-0.1
-0.4
0.0
0.6
1.1
2.0
1.7
2.2
3.5
5.0
4.4

-14.6
-31.8
-42.1
-51.7
-57.3
-65.8
-72.8
-76.3
-81.2
-90.7
-92.5
-95.8
-103.5
-107.1

-6.3
2.2
-14.5
-27.2
-47.2
-62.7
-84.9
-90.8
-96.9
-104.2
-101.1
-98.6
-102.5
-90.2

1
7
14
21
28
35
42
49
56
63
70
77
84
91

-12.8
-26.8
-37.1
-44.4
-52.1
-55.1
-61.0
-66.6
-70.8
-78.3
-79.3
-82.8
-89.9
-92.0

-9.6
-20.3
-28.7
-35.0
-39.4
-44.0
-48.4
-51.5
-54.7
-57.8
-61.3
-64.6
-65.8
-70.0

72

Table 3.8: Robustness Check: Out of Sample Events

The following table reports the OLS estimated coeﬃcients and t-statistics for fund ﬂows for out of
sample events using interaction terms to capture diﬀerences fund ﬂow sensitivity to fund character-
istics between normal and high stress periods following nationally reported traumatic events. This
table estimates diﬀerences in fund ﬂow - performance sensitivity between normal and high stress
periods. Panel (A) reports results using past year returns to assign funds into quintiles. Panel (B)
repeats the same tests using past quarter returns. Column (1) displays results for normal periods.
Column(2) shows results using only months identiﬁed as high stress periods. Column (3) reports
the diﬀerence in coeﬃcient estimates and the χ2 test under the null of no diﬀerence between the
group coeﬃcients. The variables of interest here are the reported past return quintile indicators
which estimate the convexity of the ﬂow-performance sensitivity. Return Quintile 3 is used as a
baseline for the return ranking indicatorss. Return quintile 1 groups the lowest performing funds
in each period and Quintile 5 groups the highest performing funds. Funds are ranked and assigned
quintiles separately for each CRSP fund objective identiﬁer. Year eﬀects and 3 digit CRSP objective
code eﬀects are included in each speciﬁcation. Standard errors are reported in parentheses below
the coeﬃcient estimate and are clustered at the fund-series level in every speciﬁcation. Statistical
signiﬁcance (two-sided) at the 10%, 5%, and 1% levels is denoted by *, **, and ***, respectively.

Panel A: Regression Tests

(1)

(2)

Past Year Ret. Rankings Past Year Ret. Rankings

Return Quintile 1

Return Quintile 2

Return Quintile 4

Return Quintile 5

Expense Ratio

log(Monthly Total Net Assets)

log(Fund Age)

Return Risk

log(fund family size)

Return Quintile 1 - Return Quintile 5

Year Eﬀects
Investment Style Eﬀects
Adj. R2
Num. obs.
∗∗∗p < 0.01, ∗∗p < 0.05, ∗p < 0.1

Normal Period
−0.0059∗∗∗
(−9.6136)
−0.0026∗∗∗
(−6.3206)
0.0006
(1.5103)
0.0211∗∗∗
(26.6526)
−0.1069∗
(−1.9136)
−0.0087∗∗∗
(−27.3428)
−0.0027∗∗∗
(−6.5463)
10.5537∗∗∗
(5.6477)
0.0032∗∗∗
(18.2576)
−0.027∗∗∗
(815.6)
Yes
Yes
0.0186
1039392

73

Stress Events
−0.0009
(−0.2561)
−0.0015
(−0.5444)
−0.0079∗∗∗
(−2.7314)
0.0125∗∗∗
(3.3525)
−0.0954
(−1.3241)
−0.0091∗∗∗
(−8.5187)
−0.0052∗∗∗
(−3.5937)
20.1090∗∗
(2.4365)
0.0043∗∗∗
(6.5102)
−0.0134∗∗∗
(11.615)

Yes
Yes
0.0343
23875

(3)
Diﬀ
χ2
−0.005
(2.022)
−0.001
(0.093)
0.009∗∗
(5.922)
0.009∗∗
(5.596)
0.009
(0.024)
−0.005
(0.256)
−0.001
(2.084)
−9.5553
(2.0835)
0.009
(2.370)
−0.014∗∗∗
(14.3)

Table 3.9: Fund Flows - Early vs late period

The following table reports the OLS estimated coeﬃcients and t-statistics for mutual fund ﬂows in
separate models for early and late sample sub-periods. Interaction terms are used to capture the
diﬀerence in fund-ﬂow performance convexity between normal and high stress periods. Column
(1) displays results for the 1995-2008 sub-period. Column(2) shows results for the post-2009 sub-
period. The variables of interest here are the reported interaction term coeﬃcients on the return
quintile indicators. Return Quintile 3 is used as a baseline for the return ranking indicators. Return
quintile 1 groups the lowest performing funds in each period and Quintile 5 groups the highest
performing funds. Funds are ranked and assigned quintiles separately for each CRSP fund objective
identiﬁer. Year eﬀects and 3 digit CRSP objective code eﬀects are included in each speciﬁcation.
Standard errors are reported in parentheses below the coeﬃcient estimate and are clustered at the
fund-series level in every speciﬁcation. Statistical signiﬁcance (two-sided) at the 10%, 5%, and 1%
levels is denoted by *, **, and ***, respectively.

Panel A: Regression Tests - Past Quarter Returns

Early Period 1995-2008 Late Period 2009-2016

(2)

−0.0044∗∗∗
(−4.6341)
−0.0021∗∗∗
(−3.1406)
0.0004
(0.6352)
0.0089∗∗∗
(9.6644)
−0.0087∗∗∗
(−9.4508)
0.0013
(0.8464)
0.0017
(1.2809)
0.0001
(0.0697)
−0.0066∗∗∗
(−4.5966)

Yes
Yes
Yes
Yes
0.0149
633453

Return Quintile 1

Return Quintile 2

Return Quintile 4

Return Quintile 5

Stress_Ind

Return Quintile 1:Stress_Ind

Return Quintile 2:Stress_Ind

Return Quintile 4:Stress_Ind

Return Quintile 5:Stress_Ind

Controls
Year Eﬀects
Investment Style Eﬀects
Fund Series Clustered Errors
Adj. R2
Num. obs.
∗∗∗p < 0.01, ∗∗p < 0.05, ∗p < 0.1

(1)

−0.0019∗∗
(−1.9825)
−0.0017∗∗
(−2.2196)
0.0015∗∗
(2.2671)
0.0167∗∗∗
(13.6898)
0.0036∗∗
(1.9772)
0.0063∗∗
(2.2672)
0.0037
(1.2616)
−0.0035
(−1.2967)
−0.0063∗
(−1.6863)

Yes
Yes
Yes
Yes
0.0228
429814

74

BIBLIOGRAPHY

75

BIBLIOGRAPHY

Ahern, Jennifer, Sandro Galea, Heidi Resnick, and Dean Kilpatrick, 2002, Television images and
psychological symptoms after the September 11 terrorist attacks, Psychiatry; Washington 65,
289–300.

Ahern, Kenneth R., 2018, THE IMPORTANCE OF PSYCHOLOGY IN ECONOMIC ACTIVITY:

EVIDENCE FROM TERRORIST ATTACKS, Working Paper - SSRN .

Akbas, Ferhat, Will J. Armstrong, Sorin Sorescu, and Avanidhar Subrahmanyam, 2015, Smart
money, dumb money, and capital market anomalies, Journal of Financial Economics 118, 355–
382.

American Psychological Association, 2016, Stress in America: The impact of discrimination.,

Stress in America Survey .

Antoniou, Constantinos, Anastasios Maligkris, and Alok Kumar, 2018, Terrorist Attacks, Analyst

Sentiment, and Earnings Forecasts 53.

Arin, K. Peren, Davide Ciferri, and Nicola Spagnolo, 2008, The price of terror: The eﬀects of

terrorism on stock market returns and volatility, Economics Letters 101, 164–167.

Barber, B.M., X. Huang, and T. Odean, 2016, Which Factors Matter to Investors? Evidence from

Mutual Fund Flows, Review of Financial Studies 29, 2601–2642.

Barber, Brad M., and Terrance Odean, 2000, Trading Is Hazardous to Your Wealth: The Common

Stock Investment Performance of Individual Investors, The Journal of Finance 55, 773–806.

Barber, Brad M., and Terrance Odean, 2008, All That Glitters: The Eﬀect of Attention and News on
the Buying Behavior of Individual and Institutional Investors, The Review of Financial Studies
21, 785–818.

Ben-Rephael, Azi, Shmuel Kandel, and Avi Wohl, 2010, Measuring investor sentiment with mutual

fund ﬂows, Journal of Financial Economics 104, 363–382.

Carhart, Mark M, 1997, On persistence in mutual fund performance, The Journal of ﬁnance 52,

57–82.

Chevalier, Judith, and Glenn Ellison, 1997, Risk taking by mutual funds as a response to incentives,

The Journal of Political Economy 105, 1167–1200.

Cooper, Michael J., Huseyin Gulen, and P. Raghavendra Rau, 2005, Changing Names with Style:
Mutual Fund Name Changes and Their Eﬀects on Fund Flows, The Journal of Finance 60,
2825–2858.

Cremers, KJ Martijn, and Antti Petajisto, 2009, How active is your fund manager? A new measure

that predicts performance, Review of Financial Studies 22, 3329–3365.

76

Cronqvist, Henrik, 2006, Advertising and Portfolio Choice, Ph.D., The University of Chicago, Ann

Arbor.

DellaVigna, Stefano, and Joshua M. Pollet, 2009, Investor Inattention and Friday Earnings An-

nouncements, The Journal of Finance 64, 709–749.

Drakos, Konstantinos, 2010, Terrorism activity, investor sentiment, and stock returns, Review of

Financial Economics 19, 128–135.

Edmans, Alex, Diego García, and Øyvind Norli, 2007, Sports Sentiment and Stock Returns, The

Journal of Finance 62, 1967–1998.

Elton, Edwin J., Martin J. Gruber, and Jeﬀrey A. Busse, 2004, Are Investors Rational? Choices

among Index Funds, The Journal of Finance 59, 261–288.

Foerster, Stephen, Juhani T. Linnainmaa, Brian T. Melzer, and Alessandro Previtero, 2017, Retail

Financial Advice: Does One Size Fit All?, The Journal of Finance 72, 1441–1482.

Frazzini, Andrea, and Owen A Lamont, 2008, Dumb money: Mutual fund ﬂows and the cross-

section of stock returns, Journal of Financial Economics 88, 299–322.

Gabriel, Rafael, Laura Ferrando, Enrique Sainz Cortón, Carlos Mingote, Eduardo García-Camba,
Alberto Fernández Liria, and Sandro Galea, 2007, Psychopathological consequences after a
terrorist attack: An epidemiological study among victims, the general population, and police
oﬃcers, European Psychiatry 22, 339–346.

Galea, Sandro, David Vlahov, Heidi Resnick, Jennifer Ahern, Ezra Susser, Joel Gold, Michael
Bucuvalas, and Dean Kilpatrick, 2003, Trends of Probable Post-Traumatic Stress Disorder in
New York City after the September 11 Terrorist Attacks, American Journal of Epidemiology
158, 514–524.

Goetzmann, William N., Dasol Kim, Alok Kumar, and Qin Wang, 2015, Weather-Induced Mood,

Institutional Investors, and Stock Returns, The Review of Financial Studies 28, 73–111.

Guiso, Luigi, Paola Sapienza, and Luigi Zingales, 2018, Time varying risk aversion, Journal of

Financial Economics 128, 403–421.

Hirshleifer, David, Sonya Seongyeon Lim, and Siew Hong Teoh, 2009, Driven to Distraction:
Extraneous Events and Underreaction to Earnings News, The Journal of Finance 64, 2289–
2325.

Hirshleifer, David, and Tyler Shumway, 2003, Good day sunshine: Stock returns and the weather,

The Journal of Finance 58, 1009–1032.

Israeli, Doron, Ron Kasznik, and Suhas. A Sridharan, 2017, Unexpected Distractions and Investor
Attention to Corporate Announcements, Available at SSRN: https://ssrn.com/abstract=3057278
.

Jonathan B. Berk, and Richard C. Green, 2004, Mutual Fund Flows and Performance in Rational

Markets, Journal of Political Economy 112, 1269–1295.

77

Kamstra, Mark J, Lisa A Kramer, and Maurice D Levi, 2003, Winter blues: A SAD stock market

cycle, The American Economic Review 93, 324–343.

Kamstra, Mark J., Lisa A. Kramer, Maurice D. Levi, and Russ Wermers, 2017, Seasonal Asset
Allocation: Evidence from Mutual Fund Flows, Journal of Financial and Quantitative Analysis
52, 71–109.

Kempf, Elisabeth, Alberto Manconi, and Oliver Spalt, 2017, Distracted Shareholders and Corporate

Actions, The Review of Financial Studies 30, 1660–1695.

Kuhnen, Camelia M., and Brian Knutson, 2011, The Inﬂuence of Aﬀect on Beliefs, Preferences,

and Financial Decisions, The Journal of Financial and Quantitative Analysis 46, 605–626.

Linnainmaa, Juhani T., Brian Melzer, and Alessandro Previtero, 2017, The Misguided Beliefs
of Financial Advisors, SSRN Scholarly Paper ID 3101426, Social Science Research Network,
Rochester, NY.

Lo, Andrew W., and Dmitry V. Repin, 2002, The Psychophysiology of Real-Time Financial Risk

Processing, Journal of Cognitive Neuroscience 14, 323–339.

Lo, Andrew W., Dmitry V. Repin, and Brett N. Steenbarger, 2005, Fear and Greed in Financial

Markets: A Clinical Study of Day-Traders, The American Economic Review 95, 352–359.

Lu, Yan, Sugata Ray, and Melvyn Teo, 2016, Limited attention, marital events and hedge funds,

Journal of Financial Economics 122, 607–624.

Lynch, Anthony W., and David K. Musto, 2003, How Investors Interpret past Fund Returns, The

Journal of Finance 58, 2033–2058.

MacKinlay, A. Craig, 1997, Event Studies in Economics and Finance, Journal of Economic

Literature; Nashville, Tenn. 35, 13.

Massa, Massimo, 2003, How do family strategies aﬀect fund performance? When performance-

maximization is not the only game in town, Journal of Financial Economics 67, 249–304.

Melnick, Raﬁ, and Raﬁ Eldor, 2010, Small investment and large returns: Terrorism, media and the

economy, European Economic Review 54, 963–973.

Neria, Y., A. Nandi, and S. Galea, 2008, Post-traumatic stress disorder following disasters: A

systematic review, Psychological Medicine 38, 467–480.

Neria, Yuval, Laura DiGrande, and Ben G. Adams, 2011, Posttraumatic Stress Disorder Following

the September 11, 2001, Terrorist Attacks, The American Psychologist 66, 429–446.

Odean, Terrance, 1999, Do investors trade too much?, The American Economic Review; Nashville

89, 1279–1298.

Peng, Lin, 2005, Learning with Information Capacity Constraints, The Journal of Financial and

Quantitative Analysis 40, 307–329.

78

Peress, Joel, and Daniel Schmidt, 2016, Glued to the TV: Distracted Retail Investors and Stock Mar-
ket Liquidity, SSRN Scholarly Paper ID 2814114, Social Science Research Network, Rochester,
NY.

Pfeﬀerbaum, Betty, Debby E. Doughty, Chandrashekar Reddy, Nilam Patel, Robin H. Gurwitch,
Sara Jo Nixon, and Rick D. Tivis, 2002, Exposure and peritraumatic response as predictors of
posttraumatic stress in children following the 1995 Oklahoma City bombing, Journal of Urban
Health 79, 354–363.

Pool, Veronika K., Noah Stoﬀman, and Scott E. Yonker, 2012, No Place Like Home: Familiarity

in Mutual Fund Manager Portfolio Choice, Review of Financial Studies 25, 2563–2599.

Rubin, G. James, Chris R. Brewin, Neil Greenberg, John Simpson, and Simon Wessely, 2005,
Psychological and behavioural reactions to the bombings in London on 7 July 2005: Cross
sectional survey of a representative sample of Londoners, BMJ 331, 606.

Schlenger, William E., Juesta M. Caddell, Lori Ebert, B. Kathleen Jordan, Kathryn M. Rourke,
David Wilson, Lisa Thalji, J. Michael Dennis, John A. Fairbank, and Richard A. Kulka, 2002,
Psychological Reactions to Terrorist Attacks: Findings From the National Study of Americans’
Reactions to September 11, JAMA 288, 581–588.

Schuster, Mark A., Bradley D. Stein, Lisa H. Jaycox, Rebecca L. Collins, Grant N. Marshall,
Marc N. Elliott, Annie J. Zhou, David E. Kanouse, Janina L. Morrison, and Sandra H. Berry,
2001, A national survey of stress reactions after the September 11, 2001, terrorist attacks, The
New England Journal of Medicine; Boston 345, 1507–1512.

Sensoy, Berk A., 2009, Performance evaluation and self-designated benchmark indexes in the

mutual fund industry, Journal of Financial Economics 92, 25–39.

Shaw, JON A., BROOKS Applegate, and CARYN Schorr, 1996, Twenty-One—Month Follow-up
Study of School-Age Children Exposed to Hurricane Andrew, Journal of the American Academy
of Child & Adolescent Psychiatry 35, 359–364.

Shive, Sophie, 2011, Local investors, price discovery, and market eﬃciency, Journal of Financial

Economics 104, 145–161.

Sialm, Clemens, Laura T. Starks, and Hanjiang Zhang, 2015, Deﬁned Contribution Pension Plans:

Sticky or Discerning Money?, The Journal of Finance 70, 805–838.

Silver, Roxane Cohen, E. Alison Holman, Daniel N. McIntosh, Michael Poulin, and Virginia
Gil-Rivas, 2002, Nationwide Longitudinal Study of Psychological Responses to September 11,
JAMA 288, 1235–1244.

Silver, Roxane Cohen, Michael Poulin, E. Alison Holman, Daniel N. McIntosh, Virginia Gilrivas,
and Judith Pizarro, 2004, Exploring the Myths of Coping with a National Trauma, Journal of
Aggression, Maltreatment & Trauma 9, 129–141.

Sirri, Erik R, and Peter Tufano, 1998, Costly search and mutual fund ﬂows, The journal of ﬁnance

53, 1589–1622.

79

Slone, Michelle, 2000, Responses to Media Coverage of Terrorism, The Journal of Conﬂict Reso-

lution 44, 508–522.

Szymanski, Kate, Linda Sapanski, and Francine Conway, 2011, Trauma and ADHD – Association
or Diagnostic Confusion? A Clinical Perspective, Journal of Infant, Child, and Adolescent
Psychotherapy 10, 51–59.

Terr, Lenore C., Daniel A. Bloch, Beat A. Michel, Hong Shi, and et al, 1999, Children’s symptoms
in the wake of Challenger: A ﬁeld study of distant-traumatic eﬀects and an outline of related
conditions, The American Journal of Psychiatry; Washington 156, 1536–44.

Thaler, Richard, and Cass R. Sunstein, 2008, Nudge : Improving Decisions about Health, Wealth,

and Happiness (Yale University PRess, New Haven, Conn.).

Vasterling, Jennifer J., Kevin Brailey, Joseph I. Constans, and Patricia B. Sutker, 1998, Attention

and memory dysfunction in posttraumatic stress disorder, Neuropsychology 12, 125–133.

Vasterling, Jennifer J., Lisa M. Duke, Kevin Brailey, Joseph I. Constans, Albert N. Allain, and Pa-
tricia B. Sutker, 2002, Attention, learning, and memory performances and intellectual resources
in Vietnam veterans: PTSD and no disorder comparisons, Neuropsychology 16, 5–14.

Wang, Albert, and Michael Young, 2018, Terrorist Attacks and Household Trading, Working Paper

50.

Wang, Albert, and Young, Michael, 2018, Terrorist Attacks and Investor Risk Preference: Evidence

from Mutual Fund Flows, Working Paper .

Yang, Sun, 2016, The Eﬀect of Index Fund Competition on Money Management Fees, Job Market

Paper .

Zheng, Lu, 1999, Is money smart? A study of mutual fund investors’ fund selection ability, The

Journal of Finance 54, 901–933.

80