Post on 10-Jul-2020
Electronic copy available at: https://ssrn.com/abstract=2978811
Managerial Structure and Performance-Induced Trading
Anastassia Fedyk, Saurin Patel, and Sergei Sarkissian∗
June 1, 2017
∗Fedyk is from Harvard University, Department of Economics and Harvard Business School, Boston, MA02163, USA. Patel is from the University of Western Ontario Ivey Business School, London, ON N6G0N1,Canada. Sarkissian is from the McGill University Faculty of Management, Montreal, QC H3A1G5, Canada,and Yerevan State University, Yerevan, Armenia (visiting). Fedyk may be reached at afedyk@hbs.edu. Patelmay be reached at spatel@ivey.uwo.ca. Sarkissian may be reached at sergei.sarkissian@mcgill.ca. We thankJohn Campbell and Jeremy Stein for discussions on this research project. We thank the representativesof Dodge & Cox Funds and Goldman Sachs Asset Management for sharing with us their investment deci-sion making process. The authors acknowledge �nancial support from the Social Sciences and HumanitiesResearch Council (SSHRC).
Electronic copy available at: https://ssrn.com/abstract=2978811
Managerial Structure and Performance-Induced Trading
Abstract
The literature �nds that investors increase portfolio turnover following high returns,
explaining it by either overcon�dence or skilled trading. This paper develops a theo-
retical model and shows empirically that team-managed funds trade less after good
performance than single-managed funds. The magnitude of this di�erential increases
with team size. Moreover, the change from single- to team-management structure de-
creases overcon�dence induced trading. In spite of more trading, the next-period risk-
adjusted returns of single-managed funds are no better than those of team-managed
funds. These �ndings indicate that team-management reduces overcon�dent trading.
Alternative channels cannot explain the drop in excessive trading in team-managed
funds.
JEL Classi�cation: D22; D70; G02; G23
Keywords: Behavioral bias, Excess turnover, Fund alpha, Portfolio optimization, Pos-terior belief
1
�What would I eliminate if I had a magic want? Overcon�dence.�
- Daniel Kahneman, Nobel Laureate
I. Introduction
In this paper, we study how team-management a�ects overcon�dence among mutual fund
managers. We deal with a particular type of overcon�dence: self-attribution bias, whereby
individuals tend to overly attribute outcomes, especially positive outcomes, to their own
skills. The self-attribution bias extends only to beliefs about one's own skills.1 This wedge
in attributing one's own versus others' outcomes to superior skill is the main building block
of our theoretical framework and empirical tests. Our main �nding is that team-based
managerial structure signi�cantly reduces mutual fund managers' overcon�dence as evidence
by reductions in excessive trading.
The extant literature �nds that investors often increase their portfolio turnover following
high returns, explaining this phenomenon by either overcon�dence or information-based
trading. When this excessive turnover results in some performance improvements, then
one can relate it to knowledge and skillful investing; otherwise, it is usually associated
with overcon�dence resulting from prior successes in money management (e.g., Gervais and
Odean, 2001). For example, Barber and Odean (2000, 2001), Glaser and Weber (2007),
Grinblatt and Keloharju (2009), Puetz, and Ruenzi (2011), Bailey, Kumar, and Ng (2011),
and Christo�ersen and Sarkissian (2011) all �nd that overcon�dence increases trading among
individual and professional investors without any gains in performance. Overcon�dence is
found to be detrimental in many other studies on �nancial and business decision making.2
Given such negative outcomes, understanding how �rms can mitigate overcon�dence among
1Another broad type of overcon�dence is miscalibration, when individuals overestimate the precision ofinformation (low signal volatility).
2Ben-David, Graham, and Harvey (2013) show that executives are severely miscalibrated, producing verynarrow (overcon�dent) distributions of expected returns. Scheinkman and Xiong (2003), Malmendier andTate (2005, 2008), Billett and Qian (2008), Gervais, Heaton, and Odean (2011), Malmendier, Tate, and Yan(2011), Schrand and Zechman (2012), Ahmed and Duellman (2013), and Malmendier and Taylor (2015)show that overcon�dent managers undertake suboptimal investments and value-destroying merger decisions,use less external �nance, and are prone to more intentional earnings misstatements. Camerer and Lovallo(1999) show in an experimental setting that overcon�dent people neglect the quality of their competitionand fail in business. Johnson and Fowler (2011) say that overcon�dence is prevalent among people in spiteof its many negative e�ects.
2
their decision makers is of paramount importance.3
Some studies in economics and psychology suggest that one channel through which �rms
can reduce overcon�dence in managers is collective decision-making. By working in teams,
managers avoid �bias blind spots� by catching others' behavioral mistakes, providing each
other with feedback, and thus, potentially reducing overcon�dence (Pronin, Lin, and Ross,
2002; Tetlock and Mitchell, 2009). Teams are generally smarter, more strategic, rational and
less �behavioral� than individuals which signi�cantly reduces chances of biased judgements
(Cooper and Kagel, 2005; Charness and Sutter, 2012).4 Yet, other studies �nd that group
decisions are less optimal than those of individuals.5 In particular, Sunstein and Hastie
(2014) argue that groups amplify various psychological biases, including overcon�dence, due
to incorrect informational signals from group members and reputational pressures. Therefore,
the jury is still out on how collective investment decisions a�ect the trading behavior of fund
managers.
Our theoretical model formalizes the following intuition. Individual fund managers are
subject to self-attribution bias, so that the manager proposing a particular strategy over-
attributes its success to his own skill, while other managers update their beliefs regarding
his skill rationally. Speci�cally, the model features one manager proposing a discretionary
trading strategy, and plays out over three periods. The �rst period reveals the strategy's past
3Few studies highlight positive e�ects of overcon�dence and/or downscale its extent. Galasso and Simcoe(2011) show that overcon�dent CEOs are more likely to take their �rms in a new technological direction.Gervais, Heaton, and Odean (2011) show that some overcon�dence helps managers in pursuing valuablerisky projects. Benoit and Dubra (2011) argue that much of the evidence on overcon�dence reveals only anapparent, not a true bias.
4The literature also reports the superiority of team-management for investors' performance and riskreduction. For example, Adams and Ferreira (2010) �nd that teams arrive to less extreme decisions. Sharpe(1981), Barry and Starks (1984), Sah and Stiglitz (1991), Bar, Kempf, and Ruenzi (2011) argue and �ndthat teams in the fund industry achieve diversi�cation of style and judgment that reduces portfolio risk.Empirical studies �nding a positive performance impact of teams are Hamilton, Nickerson, and Owan (2003)and Patel and Sarkissian (2016).
5For example, groups may act more aggressively and undertake riskier decisions than the average choicesof individuals in a group - phenomena known as �risky shifts� (see Wallach and Kogan, 1965; Stoner, 1968)and �group polarization� (Moscovici and Zavalloni, 1969; Sunstein, 2002). Janis (1982) develops the conceptof a �groupthink�, where people in groups accept suboptimal decisions to avoid con�icts with their colleagues.Moreover, papers, such as Prather and Middleton (2002), Chen, Hong, Huang, and Kubik, (2004), Bliss,Porter, and Schwarz (2008), Massa, Reuter, and Zitzewitz (2010), and Bar, Kempf, and Ruenzi (2011) �ndno performance di�erences between team-managed and single-managed funds. Cici (2012) �nds that whenfunds are managed by managerial teams, at the time of net out �ows they are selling substantially morewinners than losers.
3
performance, which is an imperfect signal of the manager's skill and the strategy's future
performance. In the second period, fund management decides how much capital to allocate
to the strategy going forward. The model ends in the third period, when the strategy's �nal
payo� is realized. Within this framework, single-managed funds are modeled as the single
manager, who initially proposes the strategy, deciding how much capital to allocate to it
in the second period. In team-managed funds, the second period allocation is determined
jointly by the proposing manager and his teammates.6
Coupled with any decision-making mechanism that places non-trivial weights on each
manager's opinion, this channel generates three key predictions. First, funds managed by
single individuals exhibit excessive performance-induced trading. Second, team-managed
funds exhibit less performance-induced trading than single-managed funds. Third, for team-
managed funds, performance-induced trading decreases in team size.
Our empirical tests are based on U.S. equity mutual funds data between 1991 and 2015.
Mutual fund industry is an ideal place to test our model predictions, because it provides
the largest comprehensive single source of occupational data with a rich mix of single- and
team-managed funds and a �standardized� task of generating maximum returns. In addition,
mutual fund managers perform security selection, which can be a di�cult task, and it is
precisely in such situations when people exhibit the greatest overcon�dence (Odean, 1999).
Lastly, several studies in the mutual fund literature use performance-induced trading as a
widely accepted empirical proxy for overcon�dence among fund managers which is readily
available. Following the literature, we proxy overcon�dence by performance-induced trading
which implies excessive trading by fund managers following superior performance. Figure
1 illustrates the relation between past fund performance quartiles and next period turnover
using objective-adjusted returns (OAR) of U.S. domestic equity funds.7
6Our communication with industry professionals reveals that funds are often managed by a committeevote. E�ectively, individual portfolio managers manage quasi-independent teams, where each team is re-sponsible for a speci�c strategy, but there are formal or informal cross-validation processes between theteams. For example, a more formal process might involve a junior member (not a portfolio manager) puttingtogether advocacy for a speci�c trade and proposing it to his immediate portfolio manager. If the portfoliomanager approves, then there is a more formal presentation to the senior committee with a subsequent vote.A larger group behind decision making implies that there are more people who can stop poor investmentideas or point out each others' biases.
7As the �gure shows, the fund turnover also increases after poor performance, but this increase cannot beassociated with overcon�dence; rather it results from fund managers actively changing their trading strategy
4
We begin by testing the key predictions of the model. Consistent with the model's intu-
ition, we �nd that performance-induced trading is signi�cantly lower among team-managed
funds compared to single-managed funds. That is, following strong returns, team-managed
funds increase their subsequent trading signi�cantly less than single-managed funds after
controlling for various fund and manager characteristics. This result is robust to di�erent
de�nitions of top performance (decile, quintile, or quartile) and various performance metrics,
and holds for both panel and cross-sectional regression methodologies. For example, based
on the Fama and French (2015) �ve-factor alpha and with the full set of control variables,
team-managed funds in the top quartile of performance increase their trading by about 9%
less than single-managed funds. Furthermore, consistent with another prediction of our
model, we �nd a negative relation between team size and turnover in team-managed funds
after good performance. In economic terms, two-member teams show a 7% reduction in
trading relative to single-managed funds, while funds with four or more members post a 12%
reduction in this measure.
We address the possibility of our results being driven by preexisting di�erences between
single- and team-managed funds through additional analyzes focusing on changes in fund
structure. In particular, we consider the samples of funds that switch from team- to single-
managed and vice versa, and construct matched samples of funds without changes in manage-
rial structure. The matched samples are constructed using propensity score matching based
on several fund and manager characteristics. Using the �ve closest matches, we �nd that
funds that switch from single- to team-managed see a 12% lower sensitivity of turnover to
past performance than funds that remain single-managed. Similarly, funds that switch from
team- to single-managed have 6% more performance-induced trading than their matched
counterparts that remain team-managed. In addition, we conduct placebo tests on manage-
rial structure changes and trading using middle performance quartiles and �nd no evidence
of the di�erence between team-managed and single-managed funds. This evidence helps
strengthen the causal interpretation of the relation between performance inducing trading
and team based managerial structure.
We rule out several alternative explanations that may potentially explain our results.
by replacing underperformed stocks with new securities.
5
First, we consider managerial experience. Several studies �nd that overcon�dence is per-
vasive among young and less experienced managers (e.g., Gervais and Odean, 2001; Seru,
Shumway and Sto�man, 2010; Menkho�, Schmeling and Schmidt, 2013), while others �nd
that overcon�dence may increase with experience (e.g., Heath and Tversky, 1991; Deaves,
Luders and Schroder, 2010). We show that managerial experience does not a�ect the impact
of teams on overcon�dence. Second, it is plausible that funds with superior past performance
receive higher than expected in�ows, which, in turn, force managers to trade more (e.g., Co-
val and Sta�ord, 2007; Pollet and Wilson, 2008). We address this concern by testing the
di�erential impact of net in�ows and out�ows on teams and performance-induced trading
separately. Again, we �nd no impact of fund in�ows on the relation between teams and
performance-induced trading. Third, studies show that males are more prone to overcon�-
dence than females (e.g., Lewellen, Lease, and Schlarbaum, 1977; Barber and Odean, 2001).
It is plausible then that the reduction in performance-induced trading among teams is due
to female team members. To address this concern, we test our model on male-only funds
and still �nd signi�cant reduction in performance-induced trading among male-only team-
managed funds. Finally, but very importantly, we also analyze how increased turnover a�ects
future fund performance depending on managerial structure of funds. These tests are aimed
at decoupling whether any increase in the fund turnover following superior performance is
information-based or simply re�ects overcon�dent trading. We observe that while single
managed funds trade more and, therefore, incur more costs after good returns, their next
period performance is no better and is often worse than that of team-managed funds. There-
fore, we rule out the information-based trading and conclude that single-managed funds are
more prone to overcon�dence bias.
Our paper's contribution is two-fold. First, we add to the theoretical and empirical lit-
erature that highlights the positive e�ects of �team production� on decision-making. For
example, studies show that team-based managerial structure helps overcome the problem
of e�ort coordination through peer pressure (Kandel and Lazear, 1992); diminishes extreme
and risky decisions through diversi�cation of opinions (Adams and Ferreira, 2010; Sharpe,
1981; Barry and Starks, 1984; Bar, Kempf, and Ruenzi, 2011); improves productivity and
performance (Hamilton, Nickerson, and Owan, 2003; Patel and Sarkissian, 2016). Second,
6
we contribute to the mutual fund literature that shows the prevalence of various behav-
ioral biases among fund managers. For example, studies document the disposition e�ect
(e.g., Frazzini, 2006; O'Connell and Teo, 2009; Jin and Scherbina, 2011; and Cici, 2012),
overcon�dence bias (e.g., Menkho�, Schmidt and Brozynski, 2006; Glaser and Weber, 2007;
Grinblatt and Keloharju, 2009; Puetz, and Ruenzi, 2011; Bailey, Kumar, and Ng, 2011;
and Christo�ersen and Sarkissian, 2011) and familiarity bias (e.g., Coval and Moskowitz,
1999; Pool, Sto�man and Yonker, 2012). Yet, instead of focusing on the existence of biases,
we examine the mechanism through which funds can mitigate these biases. Overall, we
interpret our �ndings as providing evidence on the bene�t of teamwork in overcoming the
overcon�dence bias among professional money managers.
The rest of the paper is organized as follows. Section 2 develops the theoretical model
of the relation between overcon�dence-induced trading and managerial structure. Section
3 describes the mutual fund data. It also presents the �rst evidence on the di�erences
in turnover response to past out-performance between single-managed and team-managed
funds. Section 4 deals with the main empirical tests. They include the examination of the
impact of managerial structure and its changes on fund turnover conditional on strong past
performance. Section 5 considers several alternative explanations for our results as well as
a series of robustness checks. Section 6 analyzes fund returns subsequent to increases in
turnover. Section 7 concludes.
II. Model of Overcon�dence and Team-Management
This section outlines a conceptual framework of decision making in mutual funds and de-
rives predictions for the relationship between trading and past performance. The key friction
in the model is overcon�dence, where an individual manager over-attributes good results to
his own skill. Due to overcon�dence, the individual manager trades more aggressively follow-
ing good past performance. Our model highlights the role of team-based decision making in
alleviating the adverse e�ects of overcon�dence: The more cautious beliefs of the manager's
teammates prevent him from overly aggressive trading. Hence, in team-managed funds, good
past performance does not induce as much subsequent trading as in individually-managed
funds.
7
A. Setup
We begin by presenting a parsimonious model of the mutual fund setting and describing
the fund managers' skills, beliefs, and objectives. The model runs over three periods, with
the time-line depicted below.
Period 1
observe signal z
+
update beliefsregarding M1's skill
Period 2
decision on how muchto trade into M1's
discretionary strategy
Period 3
payo� realized
Model time-line.
Consider a mutual fund managed by a single manager (M1) or a team of managers (M1
and Mj 6=1). The fund tracks a benchmark with return Rb. For simplicity, we assume that
Rb is a guaranteed return with no risk. In addition, the mutual fund manager M1 proposes
a discretionary trading strategy with a risky return Rd ∈ {0, R}.
The realization of the risky returnRd depends onM1's skill. In particular, the managerM1
can be either skilled or unskilled. The prior probability of a skilled manager is p. Additional
information regarding M1's skill arrives in the form of a noisy signal during period 1, and
is used by all fund managers to update their beliefs regarding the likelihood of M1 being a
skilled manager.
The signal about M1's skill is a binary outcome z ∈ {0, 1}. If the manager is skilled,
then the probability of a good signal (z = 1) is pH . If the manager is unskilled, then the
probability of a good signal is pL < pH . Thus, the joint distribution of M1's skill level and
the realization of the signal z is as follows:
(skill, z) =
(skilled, z=1) with probability p1 = ppH
(skilled, z=0) with probability p2 = p(1− pH)
(unskilled, z=1) with probability p3 = (1− p)pL(unskilled, z=0) with probability p4 = (1− p)(1− pL)
(1)
8
In the empirical analysis, we proxy for the skill signal z using past performance. In
particular, the interpretation of observing z = 1 is that the performance over the prior year
lies in the top quartile, quintile, or decile across all mutual funds.
The skill of manager M1 in�uences the future distribution of returns to his discretionary
trading. In particular, the returns from the new discretionary strategy Rd are realized in
period 3, and depend on M1's skill in the following manner. If the manager is skilled, then
the probability of a good return in period 3 is:
P{Rd = R|skilled} = q (2)
If the manager is not skilled, then the probability of a good return in period 3 is:
P{Rd = R|unskilled} = qL < q (3)
For convenience of exposition, we normalize qL = 0, which simpli�es the notation but does
not alter any of the results.
The key friction in the model is overcon�dence: the manager M1 does not correctly
update his beliefs regarding his own skill following the arrival of the signal z in period 1. In
particular, we de�ne managerial overcon�dence as follows.
De�nition 1 (Overcon�dence) Manager M1 is overcon�dent in that he overestimates the
joint probability of being skilled and receiving a good signal, and underestimates the joint
probability of being unskilled and receiving a good signal (z = 1). In particular, manager
M1 with overcon�dence parameter ∆ believes that: P{skilled, z = 1} = ppH + ∆ and
P{unskilled, z = 1} = (1 − p)pL − ∆. His beliefs conditional on a bad signal (z = 0)
are correct, as in (1).
This form of overcon�dence is consistent with a large body of empirical evidence on
the self-serving attribution bias in psychology: individuals tend to overly attribute positive
outcomes, to their own skill.8 Applications of the self-serving attribution bias to �nance
8See, for example, Langer and Roth (1975), Miller and Ross (1975), Winkler and Taylor (1979), andArkin, Appelman, and Burger (1980).
9
include short-term positive auto-correlation and long-term reversal in returns in �nancial
markets (see Daniel, Hirshleifer, and Subrahmanyam, 1998) and increased overcon�dence
among managers who have closed successful acquisition deals in the past (see Doukas and
Petmezas, 2007).
Managerial overcon�dence extends only to beliefs about one's own ability. Thus, if another
manager Mj 6=1 is present, he interprets the signal about M1's ability correctly according to
(1). This follows the extant literature in social psychology documenting the phenomenon
termed �bias blind spot": That individuals are, in general, more perceptive of others' biases
than of their own.9 The more correct beliefs about others than about oneself lead to relative
overcon�dence, i.e. more positive beliefs about oneself than about one's peers. Such relative
overcon�dence has been empirically documented in a variety of domains ranging from driving
to earning potential: the proportion of individuals who anticipate being above median is
substantially higher than 50%.10 Theoretically, relative overcon�dence and the bene�cial
e�ect of teamwork have been explored in Fedyk (2015) in the context of work assignments.
It is intuitive for a similar form of relative overcon�dence to exist in the mutual fund industry.
The remaining piece of the model setup is the investment decision. After observing the
signal and updating beliefs regarding M1's skill in period 1, the managerial team needs to
make the investment decision in period 2. Namely, the decision is how much to invest in
M1's proposed new discretionary trading strategy with stochastic return Rd, and how much
to invest in the benchmark with certain return Rb.
We model each manager's optimization problem, regardless of whether the fund is single-
managed or team-managed, as maximizing mean-variance utility in the �nal period 3 payo�,
with a risk aversion coe�cient A. In particular, if w denotes the portfolio weight on the dis-
cretionary trading strategy, then the optimization problem from the perspective of manager
i is given by:
maxw
{Ei{wRd + (1− w)Rb|z} −
A
2V ari{wRd + (1− w)Rb|z}
}, (4)
9See, for example: Pronin, Lin, and Ross (2002), Ehrlinger, Gilovich, and Ross (2005), and West, Meserve,and Stanovich (2012).
10See, for example, Svenson (1981) on relative overcon�dence about driving abilities, Weinstein (1980) onrelative overcon�dence about a variety of life events, and Alicke (1985) on overestimation of the likelihoodwith which positive adjectives are characteristic of oneself relative to one's peers.
10
subject to the no short selling constraint that the weight w ≥ 0. The expectation and
variance operators are subscripted with i to denote that the expectation and variance are
computed with respect to manager i's beliefs.
The way this optimization problem is solved by M1 in a single-managed fund is discussed
in Section 1.2 below, and the way the problem is handled jointly by {Mi}ni=1 in the team-
managed fund with n managers is described in Section 1.3. In order to ensure interior
solutions with positive discretionary trading in good circumstances, we assume that under
rational expectations, the expected return on the discretionary trading strategy conditional
on a good signal z = 1 is higher than the riskless benchmark return, i.e.:
ppHppH + (1− p)pL
qR > Rb (5)
Condition (5) simply ensures that the signal carries meaningful information � i.e., that the
optimal amount of discretionary trading is not identically equal to zero regardless of whether
the signal is good (z = 1) or bad (z = 0). The condition does not otherwise in�uence the
results.
B. Single-Managed Fund
We �rst consider the case of a single-managed fund. Here, managerial overcon�dence
induces excessive performance-based trading.
In a single-managed fund, the manager M1 solves the optimization problem (4) given his
beliefs upon observing the signal of his skill from past performance, z. In particular, given
M1's overcon�dence, his posterior beliefs are as follows:
Lemma 1 (M1's posterior beliefs) Upon observing the signal z regarding his skill, the
overcon�dent manager M1 holds the following posterior beliefs regarding the distribution of
the return Rd:
P1{Rd = 1|z = 1} =ppH + ∆
ppH + (1− p)pLq; P1{Rd = 1|z = 0} =
p(1− pH)
p(1− pH) + (1− p)(1− pL)q
(6)
E1{Rd|z} = P1{Rd = 1|z}R; V ar1{Rd|z} = P1{Rd = 1|z} (1− P1{Rd = 1|z})R2(7)
11
Proof. See Appendix.
Note that the manager's expectation of the returns to discretionary trading following a
good signal increases with his overcon�dence. Following a poor signal, however, the manager
correctly updates his beliefs regardless of his skill level. Thus, he takes too much credit for
a good signal, but correctly understands the relationship between skill and a poor signal.
As the manager factors the posterior beliefs from Lemma 1 into his optimization, he
chooses to trade more aggressively following a good signal. This is captured in the following
Proposition.
Proposition 1 (Trading in a single-managed fund) In a mutual fund managed by a
single manager with beliefs speci�ed in De�nition 1, there is more trading following a good
past performance signal (z = 1) than a poor one (z = 0), and this performance-induced
trading increases in the level of the manager's overcon�dence (∆).
Proof. See Appendix.
In our empirical setting, Proposition 1 translates to observing higher turnover following
good performance. We document this relationship in Section 4: Turnover in single-managed
funds is signi�cantly higher following performance in the top quartile, quintile, or decile than
following non-top performance. We refer to this increased turnover as �performance-induced
trading". Since we do not observe managerial overcon�dence, we do not directly test the
second part of Proposition 1; instead, we test for di�erential performance-induced trading
in single- versus team- managed funds, derived in the following subsection.
C. Team-Managed Fund
We now turn to the case of a mutual fund managed by multiple managers, and show that
the team management structure can serve to mitigate performance-induced trading.
First, we discuss the process of decision making within a team-managed fund with the
managerial team {Mi}ni=1. One of the managers, M1, proposes a discretionary trading strat-
egy with payo� Rd. The signal z is informative regarding M1's skill according to (1). Each
member of the managerial team updates his beliefs following the signal, and chooses his pre-
ferred solution to (4) according to his own beliefs. The managers then exchange their ideas
12
in an unmodeled bargaining process, and the �nal allocation to M1's discretionary trading
strategy is a weighted average of the individual managers' preferred allocations.
Formally, the team management process is de�ned as follows:
De�nition 2 (Team management process) In a fund managed by manager M1 in con-
junction with other managers Mj 6=1, the decision making process is as follows:
1. All managers observe the signal about M1's ability, z, and update their beliefs regarding
M1's skill. At the same time, M1 proposes a new discretionary trading strategy with
stochastic return Rd that depends on his skill according to equations (2)-(3).
2. Each manager Mi solves the optimization problem (4) given his updated beliefs, choos-
ing allocation weight w∗i |z. The managers discuss in an unmodeled bargaining process,
and the actual allocation weight is w∗|z =∑
i vI × (w∗|z), where vi is the weight allo-
cated by the team structure to manager Mi's preferences. All weights vi > 0.
3. As before, the payo� is realized and the game ends in period 3.
The unmodeled bargaining process in period 2 maps intuitively to a variety of natural
setups in real-life mutual funds mentioned in our conversations with industry professionals.
In some funds, the more junior portfolio managers perform the research and propose trading
strategies to the more senior team members, who give the �nal go-ahead on the trades. In
these situations, the decision weights vi favor heavily the non-proposing senior managers,
Mi 6=1; but to the extent that senior management relies on a junior manager's particular
expertise, his weight vi is also non-zero. In other funds, each manager is largely responsible
for his own asset class or strategy, but major trading decisions are approved via a majority
vote or veto by the entire group. In these situation, each manager has, at least in expectation,
a non-zero weight in the decision making process, with the realization of the weight depending
on whether his vote is pivotal. In all cases, it is reasonable to assume that in a team-managed
fund the preferences of both the strategy-proposing manager M1 and his colleagues non-
trivially enter the �nal decision, as modeled by De�nition 2.
In order to understand how the decision of the team di�ers from that of the single manager,
we highlight the way managers update their beliefs following the signal z. Recall from
13
De�nition 1 that managerM1 over-infers his own ability from a good signal z = 1. The other
managers, however, update their beliefs correctly. As a result, the implemented allocation,
which re�ects all managers' beliefs, mitigates the performance-induced trading documented
in the single-manager case. In particular, we establish the following result:
Proposition 2 (Trading in a team-managed fund) In a mutual fund managed by mul-
tiple managers, performance-induced trading is lower than in the single-managed fund. I.e.,
the di�erence in discretionary trading following a good signal (z = 1) versus a bad signal
(z = 0) is lower when the fund is team-managed.
Proof. See Appendix.
We test Proposition 2 empirically by identifying funds with single and multiple listed
managers as single- and team-managed, respectively. We observe which funds display per-
formance in the top quartile, quintile, or decile of the distribution over the previous year
(a proxy for the skill signal z), and then compare the sensitivity of turnover to the top
performance indicator for team-managed versus single-managed funds.
In order to evaluate the relation between performance-induced trading and the size of
the managerial team in team-managed funds, we make one additional assumption: that the
preferences of managers in a given team are weighted equally. In this case, the performance-
induced trading diminishes with team size. Since manager M1 is overcon�dent, his desired
allocation to his new proposed trading strategy is higher than the allocation preferred by
the other managers on the team. As the number of the other managers increases, the
equal-weighted average tilts further away fromM1's desired allocation, and the performance-
induced trading decreases. The decrease in performance-induced trading from an additional
team member is lower when the total number of managers in the team is already high. This
result can be summarized as follows:
Proposition 3 (Trading and team size) In a mutual fund managed by multiple man-
agers with equal weights, performance-induced trading monotonically decreases with the size
of the team. However, the e�ect of each additional team member becomes smaller as the
team grows.
14
We test Proposition 3 empirically by splitting the team-managed funds into those that
have two, three, and four or more managers. We compare the sensitivity of trading to past
performance in team-managed funds of each size against the sensitivity in single-managed
funds.
III. Mutual Fund Data
The mutual fund data is from MorningStar Direct. Our sample includes all domestic
U.S. equity mutual funds (excluding index and sector funds) for the 1991-2015 period. We
collect several fund and manager characteristics. The fund characteristics are: Turnover,
Size, Age, Family Size, Fees, Flows, Volatility, and two performance measures. Turnover
is the minimum of aggregated sales or aggregated purchases of securities during the year
divided by the average 12-month total net assets of the fund. Since the aggregate fund
turnover is known to change with market conditions and investment style, we also compute
excess turnover. Fund Size (in millions of dollars) is the total net assets under management
of a fund in a given year. Fund Age (in years) is the di�erence between the fund's inception
year and the current year. Family Size (in billions of dollars) is measured by the total net
assets under management of the fund complex to which the fund belongs at the end of the
calendar year. Fund Fees (percent per year) is the annual total expense ratio of the fund.
Return Volatility (percent per year) is the standard deviation of monthly gross fund returns
over the past 12 months. Flows is de�ned as the net growth in the total net assets of a
fund, as a percentage of the fund's total net assets, adjusted for the prior year return. To
reduce the in�uence of outliers, we trim Turnover and Excess Turnover, Fund Fees, and Fund
Flows at 1% and 99% levels. The two performance measures are: OAR, which is the annual
objective-adjusted gross fund return computed each year and α(FF5), which is the alpha
based on the Fama and French (2015) �ve-factor model that adds pro�tability (RMW) and
investment (CMA) factors to the Fama-French three factor model. Finally, we also compute
two managerial characteristics. The �rst is Manager Industry Tenure (in years), which is the
number of years the fund manager has been within the fund industry. The second is Female,
which is de�ned as the proportion of female managers in a fund.
Table 1 reports the summary statistics. Panel A shows the distribution of single-managed
15
and team-managed funds during the sample period and across di�erent managerial struc-
tures. There are more than 38,000 fund-year observations, and the number of observations
for team-managed funds is about 70% larger than single-managed funds. As documented
in earlier studies (see Patel and Sarkissian, 2016), the percentage of single-managed funds
has drastically decreased over time from almost 71% in 1991 to below 29% by 2015. Conse-
quently, the proportion of team-managed funds has increased, and the largest hike is observed
among teams consisting of four and more managers - almost six-fold from 4% to 23%.
Panel B of Table 1 reports the number of observations, means, and the standard deviation
of fund and manager characteristics of single-managed and team-managed funds. We can see
that on average, single-managed funds have signi�cantly larger turnover than team-managed
funds (86% versus 78%). Single-managed funds also are signi�cantly smaller, younger, and
are managed by managers with shorter industry tenure than team-managed funds. How-
ever, family size where single-managed funds operate is larger than that of team-managed
funds. Finally, single-managed funds on average outperform team-managed funds in terms
of objective-adjusted returns and the Fama-French �ve-factor alphas.
To give the �rst illustration on how fund managerial structure impacts performance-
induces trading in Figure 2 we show the relation between past fund OARs and subsequent
turnover for single-managed funds and team-managed funds of di�erent sizes. Plot A com-
pares the relation between single-managed funds and all team-managed funds. We can see
that on average the turnover of team-managed funds is lower than that of single-managed
funds across all performance quartiles. More importantly, we observe a divergence of turnover
for the highest return quartile, that is, single-managed funds increase their turnover following
strong performance much more than team-managed funds. Plots B, C, and D demonstrate
the same relation between single-managed and team-managed funds with two, three, and
four plus members, respectively. We can see that the di�erence between increased trad-
ing activity following previous out-performance between single-managed and team-managed
funds seems to be increasing with team size. This is generally consistent with the empirical
implications of our model and, in the absence of the next period gains, can well be explained
with overcon�dence resulting from past successful investments.
Figure 3 repeats the plots from Figure 2 but this time using the Fama-French �ve-factor
16
alphas as the fund performance measure. Again, Plot A deals with single-managed funds
and all team-managed funds, while Plots B, C, and D consider team-managed funds of
di�erent sizes. The patterns are similar to those in the corresponding plots in Figure 2.
However, we note that the turnover di�erence between single- and team-managed funds
for high-performing funds is larger than in Figure 2, while that for low-performing funds
(Quartile 1) is smaller or non-existent at all, as in the case of Plot C with three-member
teams. Therefore, Figures 2 and 3 document substantial turnover di�erences following strong
returns between single-managed and team-managed funds. In the next section, this �nding
is tested statistically in more rigorous empirical settings.
IV. Empirical Tests
A. Managerial Structure and Turnover
Our �rst set of empirical tests analyzes the impact of the current managerial structure of
a fund on its turnover. To properly account for genuine di�erences in fund trading intensity
over our sample period and across fund types, our main dependent variable is the excess
turnover, Turnoverex (percent per year). It is de�ned as the di�erence between the fund
turnover in a given year and the median turnover for all funds with the same fund investment
objective in that year. Since overcon�dence in �nancial markets is associated with increased
trading following superior performance, to examine the main empirical predictions of our
theoretical model, we use the following regression speci�cation:
Turnoverexi,t = β0 + β1Teami,t−1 × Perfi,t−1 + β2Perfi,t−1 + β3Teami,t−1
+ β4Controlsi,t−1 + β5FEi + ei,t, (8)
where Teami,t−1 is a dummy variable, which is equal to one if fund i is team managed
at the end of year t − 1, and zero otherwise, Perfi,t−1 is a dummy variable, which is equal
to one if fund i has the top performance rank, and zero otherwise. Our main performance
metric is the Fama-French �ve-factor alpha. We consider three top performance cut-o�s:
17
Top 25%, top 20%, and top 10%. Controlsi,t−1 is the set of control variables, which includes
all fund and manager characteristics from Panel (B) of Table 1; FEi are the fund �xed
e�ects, which include fund investment objective �xed e�ects and fund family �xed e�ects,
since average trading intensity is known to vary across funds with di�erent investment goals
and fund family culture. We also control for fund location with the inclusion of the �nancial
city indicator variable following Christo�ersen and Sarkissian (2011). Finally, we cluster
standard errors by fund and year.
Table 2 shows the results from estimating model (8) for three di�erent cut-o�s of the top
fund performance: Columns 1-3 for the top quartile performance, columns 4-6 for the top
quintile performance, and columns 7-9 for the top decile performance. The regressions in
columns 1, 4, and 7 do not include fund and manager controls. Intercept is included in each
regression but its estimates are not reported. Across all these estimations the coe�cient
on Perfi,t−1 is positive and signi�cant con�rming Proposition 1 of the model that single-
managed funds substantially increase trading following good returns. We also �nd that both
non-interactive and interactive Team coe�cients are negative and signi�cant, implying that
team-managed funds engage in much less trading overall, and especially after experiencing
strong prior year returns. The bottom of the table shows the di�erence in the coe�cients
between team-managed and single-managed funds, Di� (T-S), and the p-value of the corre-
sponding F-test. It reveals that the excess turnover of team-managed funds following their
top performance is about 13% lower than that of similarly performing single-managed funds,
as predicted by Proposition 2 of the model. This di�erence is statistically signi�cant at the
1% level across all three estimations. In fact, the share of this lower di�erence attributed
directly to funds being in the top performance bracket is about 8-9%, as one can infer from
the magnitude of the slope coe�cient on the term.
Regressions 2, 5, and 8 include the full set of controls. This inclusion reduces our sample
size substantially, by about 22%. In spite of this reduction in the number of observations,
we still �nd a negative and signi�cant coe�cient for the Team variables, especially on the in-
teractive term. The di�erence tests again show that team-managed funds trade signi�cantly
less than their single-managed peers following top performance results in the prior year. In
this case the economic magnitude of the di�erence ranges between 10% (using top 20% per-
18
formance) and 13% (using top 10% performance). In the last set of regressions (columns 3, 6,
and 9) we also add the fund family �xed e�ects. This alteration does not materially change
the observed relation between team-management and lower turnover subsequent to top fund
performance even further, and, if anything, the result becomes marginally stronger. Now
the di�erence tests show that the impact of team-management on reducing fund turnover
among last year's top decile performers exceeds 16% (column 9).
Thus, Table 2 shows that after achieving much better than average returns, team-managed
funds exhibit a signi�cantly lower propensity to subsequently increase their portfolio turnover
than single-managed funds, consistent with Propositions 1 and 2 of our theoretical model.
This result re�ects not only the lower trading activity of team-managed funds in general
as shown in Table 1, but, more importantly, the less excited reaction of these types of
funds to their past superior performance. Note that the negative relationship between team-
management and fund turnover conditional on performance can be associated not only with
overcon�dence but also with informed trading. If single-managed funds have better investing
skills, they may increase their trading activity following strong performance. We address this
issue later in the paper, showing that the single-managed funds' increased turnover does not
translate to better future performance.
Since our conclusions are immune to top fund performance cut-o� percentiles, in all
subsequent estimations we use the top quartile level. This choice implies a larger sample size
for our performance variable, increasing the power of the corresponding test statistics.
In Table 3 we repeat our estimations from Table 2 for the top performance quartile across
managerial teams of di�erent sizes. To conserve space, we report only the coe�cients and
their p-values for the team and performance variables. The last two lines of the table again
show the di�erence in the coe�cients between team-managed and single-managed funds and
its signi�cance. Columns 1-3 compare two-member team funds with single-managed funds,
columns 4-6 compare three-member team funds with single-managed funds, and columns
7-9 report a comparison of funds with four or more members against single-managed funds.
Across all regression speci�cations (without and with controls, and without and with fund
family �xed e�ects) and in spite of much smaller sample sizes, the excess turnover of team-
managed funds is signi�cantly lower than that of their single-managed counterparts irrespec-
19
tive of the team size. The only exception to this evidence is the statistically insigni�cant
di�erence in column 5, when three-manager funds are compared against single-managed ones
without accounting for fund family e�ects. Also, note that the coe�cients on the interac-
tive term, Teami,t−1 × Perfi,t−1, in the most comprehensive regressions for three and four
and more manager funds in columns 6 and 9, respectively, are about twice as large in mag-
nitude as the corresponding slope in column 3 for two-manager teams. This implies that
larger teams decrease the likelihood of excessive trading after good performance even more
than a managerial team composed of only two people, as suggested by Proposition 3 of our
theoretical model.
B. Changes in Managerial Structure and Fund Turnover
One limitation of the previous test results is that, due to the persistence in managerial
structure of funds and the three-year window metrics for our performance measures, they
could possibly be related to other fund di�erences not captured by the speci�cation of model
(1). A cleaner test would examine the performance-turnover relation immediately following
the changes in funds' managerial structure. The expectation is that the move to team-
management should decrease the likelihood of excessive trading after good performance,
while the move to single-management should increase it. To perform this test, we identify
all instances of funds that change their managerial structure during our sample period. There
are 233 funds that switch from single- to team-management and 162 funds that switch from
team- to single-management.
We repeat our model (8) estimations solely using fund-year observations for these two
very small sub-samples of the data and present the results in Table 4. In this table, we
replace the Team dummy with the change in managerial structure (∆MS) dummy. The �rst
three columns show the estimates in cases of funds becoming team-managed in year t-1. In
these cases, ∆MS is a dummy that takes the value of one if a fund is team-managed at time
t-1 but was single-managed at time t-2. We can see economically and, for the most part,
statistically signi�cant coe�cients on the interactive ∆MS term, similar to the results in
Table 2. The formal F-test on the di�erences in sensitivity of turnover to past performance
between team-managed and single-managed funds is signi�cant across all three regressions
20
with the weakest 10% level recorded in column 3. Note that these di�erences are again
economically comparable to those in Table 2 and their slightly weaker statistical signi�cance
re�ects solely the drastic sample size reduction in these estimations.
The last three columns of Table 4 show the estimates in cases of funds becoming single-
managed in year t− 1. For these estimations, ∆MS is a dummy that takes the value of one
if a fund is single-managed at time t − 1 but was team-managed at time t − 2. Now, the
coe�cient on the ∆MSi,t−1 × Perfi,t−1 term is large and positive in columns 5 and 6 � the
two most comprehensive regression speci�cations. The coe�cients and the di�erences are
insigni�cant statistically, again primarily because of the small sample size. However, in spite
of the loss of statistical power associated with much smaller samples relative to those in Table
2, the central message of our tests in Table 4 is clear � a change in the managerial structure of
funds has a large in�uence on how actively funds trade following good performance. In these
cases, the move towards team-management unambiguously reduces a fund's excess turnover.
To further re�ne our analysis, we proceed to compare the trading behavior of funds that
undergo changes in their managerial structure (treated sample) with otherwise similar funds
that see no alterations to their management (control sample). To create the control sample,
we apply the propensity score matching methodology using logistic regressions to identify
funds that share similar observable fund characteristics with the treatment group. Each fund
that switches its managerial structure is matched, with replacement, to one or more funds
with the closest propensity scores based on fund characteristics such as performance, size,
age, �ows, expenses, family size, and investment objective, as well as manager characteristics
such as industry tenure and gender composition in the same period. Then we calculate the
di�erence in excess turnover of funds that switch from single (team) to team (single) and
funds that are in the matched sample.
Table 5 shows how changes in managerial structure of funds in�uence the follow up excess
turnover after recording top 25% performance using the treated and matched fund samples.
The �rst three columns report the results for funds that become team-managed at time t-1.
The last three columns report the results for funds that become single-managed at time t-1.
The regression speci�cations are analogous to those in Table 4, and, as before, we do not
report the estimates of control variables and the intercept. Panel A shows the estimation
21
results for the one-to-one fund matching scheme. It leads to 466 observations for changes from
single- to team-management and 324 changes from team- to single-management, including
both the treated and the matched samples. We �nd a negative, very large in magnitude, and
statistically highly signi�cant coe�cient on the treated funds sample. In economic terms,
based on the regression with all �xed e�ects (column 3), the change to team structure leads
to a 21% drop in excess turnover in comparison to the control group of funds. On the other
side, the move towards single-management again increases excess turnover. The magnitude
of the treated group coe�cient is between 5% and 7%, depending on the speci�cation.
Panel B of Table 5 shows the estimation results for the one-to-�ve fund matching scheme.
In this case the samples are expectantly larger, as we have 792 observations for changes
from single- to team-management and 553 from team- to single-management, including the
matched samples. In this case, we again see that the slope coe�cients associated with the
treated funds are negative and highly signi�cant, albeit with relatively smaller magnitudes
than the corresponding estimates in Panel A. Also, similar to the output in Panel A, the
tests examining switches from team- to single-management produce consistently positive and
economically sizable coe�cients on the treated funds. In addition, due to the larger sample
sizes, now we also reach marginal signi�cance levels in columns 4 and 5.
Finally, we want to illustrate that the behavior of our samples of treated and control
groups in impacting the excess turnover documented in Table 5 is indeed limited to the top
performance funds. We can achieve this by considering placebo tests of the same scenarios as
in Table 5 but for middle performance quartiles. Table 6 shows the results of such tests. In
Panel A, both treated and control groups consist of funds in the second performance quartile
(between 25th and 50th percentiles). In Panel B, both fund groups consist of funds in the
third performance quartile (between 50th and 75th percentiles). All other speci�cations
and panel formats are the same as in Table 5. The results across both panels for changes
in the managerial structure from single- to team-management show that coe�cients on the
treated sample are not only statistically but also economically insigni�cant. Generally similar
patterns are observed in the cases of funds switching from team- to single-management, with
one marginal exception for funds in the third performance quartile: In column 6, based on
the full regression speci�cation with all �xed e�ects, the coe�cient on the treated funds is
22
relatively large (12%) and marginally signi�cant in spite of only 338 observations. Note that
this outcome is still consistent with the overall logic, since funds in the third performance
quartile (above the median) are considered as better performing institutions.
C. Cross-Sectional Tests
The literature documents that panel and cross-sectional regressions may produce point
estimates that di�er across estimators (see Sul, 2016). All of our previous tests are conducted
in panel regressions, which may have some issues arising from cross-sectionally correlated
standard errors even after accounting for clustering across funds and time (e.g., Pesaran,
2006). Therefore, it is important to illustrate that our �ndings hold irrespective of the
estimation methodology.
In Table 7, we show the results from standard Fama-Macbeth cross-sectional tests, where
the dependent variable is again excess turnover of fund i during each year t. All fund and
manager controls, as well as �xed e�ects, are the same as in previous panel tests. Estimation
in column 1 does not include control variables. Column 1 and 2 includes only objective �xed
e�ects. The full model speci�cation with the inclusion of all controls and both objective and
fund family �xed e�ects is presented in column 3. We again report only the main coe�cients
of interest. Overall, our results are similar to those in Table 2. Across all three speci�cations
the coe�cient on Teami,t−1 × Perfi,t−1 is negative and signi�cant at least at the 5% level,
implying that team-management reduces excess turnover resulting from previously reported
strong fund performance. Note that the inclusion of all controls and �xed e�ects increases
this coe�cient in magnitude. The last two rows of the table again show the di�erence in
excess turnover between team-managed and single-managed funds, as well as the p-value of
the corresponding F-test. These numbers are again very similar to the corresponding values
in Table 2.
V. Alternative Explanations and Speci�cations
In this sub-section we consider a range of alternative explanations that could potentially be
related to our evidence on the lower sensitivity of turnover to good past performance among
team-managed funds. The e�ects considered below could have a convoluted linkage with
23
team structure and, therefore, impact the estimation of the importance of team-management
for the performance-turnover relation.
A. Manager Experience
The �rst alternative is fund manager age and/or experience. Many studies �nd that over-
con�dence is especially pervasive among younger and less experienced people (e.g., Gervais
and Odean, 2001; Seru, Shumway and Sto�man, 2010; Christo�ersen and Sarkissian, 2011;
Menkho�, Schmeling, and Schmidt, 2013). However, other papers show that under certain
conditions, overcon�dence may also increase with age and experience (e.g., Heath and Tver-
sky, 1991; Deaves, Luders, and Schroder, 2010). Therefore, it is possible that our results
are driven largely by funds with certain management structures disproportionately tilting
towards either less or more experienced portfolio managers.
To address this concern, in columns 1 and 2 of Table 8, we split our sample by the fund
managers' experience level, considering separately funds whose managers have an average
of less than ten years of industry experience and those with average managerial experience
of ten years or more. In team-managed funds the industry experience is measured as the
average experience of all managers in a given fund. We then rerun our model (1) tests
similar to those in Table 2 with the full set of control variables and �xed e�ects. As we
can see, the coe�cient estimates on the interactive term, Teami,t−1 × Perfi,t−1, are negative
and signi�cant for funds managed by both less experienced and more experienced fund
managers. Moreover, the di�erence tests between team-managed and single-managed funds
at the bottom of the table for both sub-samples are again highly signi�cant. Among control
variables that increase turnover the most notable ones are the volatility and the �nancial
city dummy for less experienced managers, the latter being consistent with earlier results in
Christo�ersen and Sarkissian (2011).
B. Changes in Fund Flows
The second plausible alternative is that changes in trading intensity simply re�ect changes
in �ows to mutual funds. For example, Pollet and Wilson (2008) �nd that in response to
net fund in�ows, funds usually increase their investments in existing holdings. Then, top
24
performing team-managed funds may increase their turnover following good returns to a
lesser extent than single-managed funds due to weaker in�ows of investors' money rather
than due to lower overcon�dence. We address this concern in columns 3 and 4 of Table 8,
where we split our sample into funds with net in�ows and net out�ows. The coe�cient on
the interactive Team term is negative in both estimations, but it is much larger in magnitude
and statistically highly signi�cant only in the case of net out�ows. In addition, in the case
of in�ows, fund managers have a choice between immediately investing new money and
keeping it for a while in cash. Yet, in case of out�ows, funds are forced to reduce their
holdings immediately. Therefore, net investor �ows cannot explain our results.
C. Tournament Behavior
The third alternative explanation is based on mutual fund tournaments. For example,
Brown, Harlow, and Starks (1996) show that funds strategically shift risk levels by increasing
the volatility of their portfolios in the second half of the calendar year when they underper-
form in the �rst half to attract additional fund �ows. As a result, funds that engage in this
strategic risk-shifting end up trading signi�cantly more than funds that choose not to shift
risk (Huang, Sialm, and Zhang, 2011). Then, one may argue that a higher portfolio turnover
that we observe in single- managed funds compared to team-managed funds arises because
single-managed funds are more likely to engage in risk-shifting rather than trade as a result
of overcon�dence. To control for the impact of changes in risk levels on turnover, we add
the absolute change in volatility as a new explanatory variable in our regression speci�cation
in column 5 of Table 8. A positive coe�cient on absolute volatility implies that changes in
fund return volatility increase funds' portfolio turnover. This is indeed what we �nd � the
coe�cient on absolute change in volatility is positive. However, this has very little impact
on our coe�cient of interest, Teami,t−1 × Perfi,t−1, which is still negative and statistically
signi�cant in both economic and statistical terms. Therefore, the di�erences in portfolio
turnover after top past performance of single- and team-managed funds cannot be explained
by changes in risk levels due to tournament incentives.
25
D. Gender
Fourth, it is also possible that teams lead to less excessive trading after good performance
because larger manager groups have higher likelihood of including females, which are known
to be less overcon�dent than males (e.g. see Lewellen, Lease, and Schlarbaum, 1977; Bar-
ber and Odean, 2001). Therefore, in column 6 of Table 8, we re-estimate our full model
(1) speci�cation for male-only single- and team-managed funds. We still observe that the
interactive team coe�cient and the di�erence between team- and single-managed funds are
negative and statistically highly signi�cant. This implies that gender composition cannot
explain our results either. Thus, team-management itself appears to be the major driver for
the reduction of propensity of mutual fund managers to increase their trading after strong
performance.
E. Robustness Issues
Finally, in Table 9 we repeat our main panel tests on three alternative fund performance
metrics, namely: The objective-adjusted return, OAR, the Carhart (1997) four-factor alpha,
α(C4), and the Pastor and Stambaugh (2003) alpha, α(PS), that is computed from the
Cahrart (1997) model by adding to it the liquidity factor of Pastor and Stambaugh (2003).
In the last column of the table we also report the estimates using the Fama-French �ve-
factor alpha, α(FF5), but under alternative �xed e�ect settings � with fund �xed e�ects and
interactive objective and time �xed e�ects. The estimations for each of the other performance
metrics are conducted with and without family �xed e�ects. For brevity we report only the
di�erence in excess turnover between team-managed and single-managed funds for the top
25% performance quartile as in Table 2 as well as the p-value of the corresponding F-test.
We observe a consistent pattern across all estimation results, con�rming our earlier �nding:
Team-managed funds trade signi�cantly less than single-managed funds following strong
performance in the prior year. Even the point estimates of the di�erence are very similar
across almost all performance measures and model speci�cations with the widest di�erence
archived for the Fama-French �ve-factor alpha, when the regression includes interactive fund
investment objective and time �xed e�ects.
26
VI. Impact on Future Performance
As we mentioned earlier, funds may increase trading activity after posting strong returns
not only because of overcon�dence but also because of better investment knowledge. That
is, excess turnover would not be considered harmful if it leads to out-performance in the
subsequent period. In our setting, this implies that if single-managed funds increase their
turnover and report better returns than team-managed funds over the following period, then
at least a sizable part of their higher trading could be attributed to their managers' skills.
Note from Table 1 that single-managed funds have higher, on average, OARs than team-
managed funds, and there are more single-managed funds within larger fund families. It may
be that skillful single-managed funds with high turnover from the most established players
in the mutual fund industry are responsible for this average performance di�erence.
We examine the above issue directly in Table 10, which shows the relation between excess
turnover and future fund performance for single-managed and team-managed funds condi-
tional on past fund returns. We use all four performance metrics as our dependent variables.
Panel A conditions on the top quartile of past fund performance, while Panel B focuses on
the top decile. The main independent variable of interest is excess turnover. All fund and
manager controls are the same as in Table 2. Fixed e�ects include fund investment objective
times year �xed e�ects, as well as fund family �xed e�ects. The most important results we
observe in both panels is that the sign on excess turnover is negative for all risk-adjusted
returns irrespective of the managerial structure of funds. For OARs it is positive in both
panels but insigni�cant. Therefore, the �rst conclusion is that additional turnover following
strong prior performance is not predictive of another year of superior returns.
More importantly, however, is that the point estimates on the turnover term for single-
managed funds are lower than those for team managed funds for every risk-adjusted return
measure. While this di�erence is not statistically signi�cant (we do not report it explic-
itly), in economic terms the average spread in the trading impact on α(C4), α(PS), and
α(FF5) is 0.0320 in Panel A and 0.0229 in Panel B. This means that a 100% increase in
the current year trading of high-performing team-managed funds reduces their next period's
risk-adjusted return by 0.023-0.032 percent per month (28-39 basis points per year) less
27
than a similar trading increase of high-performing single-managed funds. In addition, the
only highly signi�cant (negative) impact of excess turnover on future return is recorded with
single-managed funds based on the α(PS) measure and top decile performance. Thus, we
conclude that the extra trading among single-managed funds relative to their team-managed
counterparts observed after top performance in the previous period is not bene�cial and
even damaging for their subsequent performance. This result provides additional empirical
evidence in support of our model that team structure of portfolio management in mutual
funds reduces overcon�dent and harmful trading.
VII. Conclusion
In this paper we establish a theoretical relation between the organizational structure of
mutual funds and the likelihood of their managers engaging in overcon�dent (excessive) trad-
ing following superior performance. Our model implies that team-managed funds exhibit less
overcon�dence than single-managed funds. Subsequent empirical tests support the model
predictions. Moreover, we show that the model predictions hold also when funds undergo
changes in their managerial structure. In particular, a shift from single- to team-management
signi�cantly decreases out-performance-induced trading, while a shift from team- to single-
management leads generally to the opposite result. Our �ndings are robust to various cut-o�s
of top fund performance and to the presence of both fund-speci�c and manager-level control
variables. They are also immune to fund performance metrics and other econometric alter-
ations, such as panel and cross-sectional regression methodologies. Moreover, we show that
our main �ndings cannot be accounted for by a range of potential alternative explanations.
28
References
Adams, R., and D. Ferreira, 2010, Moderation in groups: Evidence from betting on ice
break-ups in Alaska, Review of Economic Studies 77, 882-913.
Ahmed, A., and S. Duellman, 2013, Managerial overcon�dence and accounting conser-
vatism, Journal of Accounting Research 51, 1-30.
Alicke, M., 1985, Global self-evaluation as determined by the desirability and controlla-
bility of trait adjectives, Journal of Personality and Social Psychology 49, 1621-1630.
Arkin, R., A. Appelman, and J. Burger, 1980, Social anxiety, self-presentation, and the
self-serving bias in causal attribution, Journal of Personality and Social Psychology 38, 23-
35.
Bailey, W., A. Kumar, and D. Ng, 2011, Behavioral biases of mutual fund investors,
Journal of Financial Economics 102, 1-27.
Bar, M., A. Kempf, and S. Ruenzi, 2011, Is a team di�erent from the sum of its parts?
Evidence from mutual fund managers, Review of Finance 15, 359-396.
Barber, B., and T. Odean, 2000, Trading is hazardous to your wealth: The common stock
investment performance of individual investors, Journal of Finance 55, 773-806.
Barber, B., and T. Odean, 2001, Boys will be boys: Gender, overcon�dence, and common
stock investment, Quarterly Journal of Economics 116, 261-292.
Barry, C., and L. Starks, 1984, Investment management and risk sharing with multiple
managers, Journal of Finance 39, 477-491.
Ben-David, I., J. Graham, and C. Harvey, 2013, Managerial miscalibration, Quarterly
Journal of Economics 128, 1547-1584.
Benoit, J., and J. Dubra, 2011, Apparent overcon�dence, Econometrica 79, 1591-1625.
Billett, M., and Y. Qian, 2008, Are overcon�dent CEOs born or made? Evidence of
self-attribution bias from frequent acquirers, Management Science 54, 1037-1051.
Bliss, R., M. Porter, and C. Schwarz, 2008, Performance characteristics of individually-
managed versus team-managed mutual funds, Journal of Portfolio Management 34, 110-119.
Brown, K., W. Harlow, and L. Starks, 1996, Of tournaments and temptations: An analysis
of managerial incentives in the mutual fund industry, Journal of Finance 51, 85-110.
29
Camerer, C., and D. Lovallo, 1999, Overcon�dence and excess entry: An experimental
approach, American Economic Review 89, 306-318.
Carhart, M., 1997, On persistence in mutual fund performance, Journal of Finance 52,
57-82.
Charness, G., and M. Sutter, 2012, Groups make better self-interested decisions, Journal
of Economic Perspectives 26, 157-176.
Chen, J., H. Hong, M. Huang, and J. Kubik, 2004, Does fund size erode mutual fund
performance? The role of liquidity and organization, American Economic Review 94, 1276-
1302.
Christo�ersen, S., and S. Sarkissian, 2011, The demographics of fund turnover, Journal
of Financial Intermediation 20, 414-440.
Cici, G., 2012, The prevalence of the disposition e�ect in mutual funds' trades, Journal
of Financial and Quantitative Analysis 47, 795-820.
Cooper, D., and J. Kagal, 2005, Are two heads better than one? Team versus individual
play in signaling games, American Economic Review 95, 477-509.
Coval, J., and T. Moskowitz, 1999, Home bias at home: Local equity preference in do-
mestic portfolios, Journal of Finance 54, 2045-2073.
Coval, J., and E. Sta�ord, 2007, Asset �re sales (and purchases) in equity markets, Journal
of Financial Economics 86, 479-512.
Daniel, K., D. Hirshleifer, and A. Subrahmanyam, 1998, Investor psychology and security
market under and overreactions, Journal of Finance 53, 1839-1885.
Danz, D., K. Madarasz, and S. Wang, 2014, Information projection: An experimental
investigation, Working paper, London School of Economics.
Deaves, R., E. Luders, and M. Schroder, 2010, The dynamics of overcon�dence: evidence
from stock market forecasters, Journal of Economic Behavior and Organization 75, 402-412.
Doukas, J., and D. Petmezas, 2007, Acquisitions, Overcon�dent Managers and Self-
attribution Bias, European Financial Management 13, 531-577.
Ehrlinger, J., T. Gilovich, and L. Ross, 2005, Peering into the bias blind spot: People's
assessments of bias in themselves and others, Personality and Social Psychology Bulletin 31,
680-692.
30
Fama, E., and K. French, 2015, A �ve-factor asset pricing model, Journal of Financial
Economics 116, 1-22.
Fedyk, A., 2015, Overcoming overcon�dence: Teamwork and self-control, Working paper,
Harvard University.
Frazzini, A., 2006, The disposition e�ect and underraction to news, Journal of Finance
61, 2017-2046.
Galasso, A., and T. Simcoe, 2011, CEO overcon�dence and innovation, Management
Science 57, 1469-1484.
Gervais, S., J. Heaton, and T. Odean, 2011, Overcon�dence, compensation contracts, and
capital budgeting, Journal of Finance 66, 1735-1777.
Gervais, S., and T. Odean, 2001, Learning to be overcon�dent, Review of Financial Studies
14, 1-27.
Glaser, M., and M. Weber, 2007, Overcon�dence and trading volume, Geneva Risk and
Insurance Review 32, 1-36.
Grinblatt, M., and M. Keloharju, 2009, Sensation seeking, overcon�dence, and trading
activity, Journal of Finance 64, 549-578.
Hamilton, B., J. Nickerson, and H. Owan, 2003, Team incentives and worker heterogene-
ity: An empirical analysis of the impact of teams on productivity and participation, Journal
of Political Economy 111, 465-497.
Heath, C., and A. Tversky, 1991, Preference and belief: ambiguity and competence in
choice under uncertainty, Journal of Risk and Uncertainty 4, 5-28.
Huang, J., C. Sialm, and H. Zhang, 2011, Risk Shifting and Mutual Fund Performance,
Review of Financial Studies 24, 2575-2616.
Jin, L., and A. Scherbina, 2011, Inheriting loser, Review of Financial Studies 24, 786-820.
Janis, I., 1982, Groupthink: Psychological studies of policy decisions and �ascoes, Second
Edition, New York: Houghton Mi�in.
Johnson, D., and J. Fowler, 2011, The evolution of overcon�dence, Nature 477(7364),
317-320.
Kandel, E., and E. Lazear, 1992, Peer pressure and partnerships, Journal of Political
Economy 100, 801-817.
31
Langer, E., and J. Roth, 1975, Heads I win tails it's chance: The illusion of control as
a function of the sequence of outcomes in a purely chance task, Journal of Personality and
Social Psychology 32, 951-955.
Lewellen, W., R. Lease, and G. Schlarbaum, 1977, Patterns of investment strategy and
behavior among individual investors, Journal of Business 50, 296-333.
Malmendier, U., and G. Tate, 2005, CEO overcon�dence and corporate investment, Jour-
nal of Finance 60, 2661-2700.
Malmendier, U., and G. Tate, 2008, Who makes acquisitions? CEO overcon�dence and
the market's reaction, Journal of Financial Economics 89, 20-43.
Malmendier, U., G. Tate, and J. Yan, 2011, Overcon�dence and early-life experiences:
The e�ect of managerial traits on corporate �nancial policies, Journal of Finance 66, 1687-
1733.
Malmendier, U., and T. Taylor, 2015, On the verges of overcon�dence, Journal of Eco-
nomic Perspectives 29, 3-8.
Massa, M., J. Reuter, and E. Zitzewitz, 2010, When should �rms share credit with employ-
ees? Evidence from anonymously managed mutual funds, Journal of Financial Economics
95, 400-424.
Menkho�, L., M. Schmeling, and U. Schmidt, 2013, Overcon�dence, experience and pro-
fessionalism: an experimental study, Journal of Economic Behavior and Organization 86,
92-101.
Miller, D., and M. Ross, 1975, Self-serving Biases in the attribution of causality: Fact or
Fiction, Psychological Bulletin 82, 213-225.
Moscovici, S., and Zavalloni, M., 1969, The group as a polarizer of attitudes. Journal of
personality and social psychology 12, 125-131.
O'Connell, P., and M. Teo, 2009, Institutional investors, past performance, and dynamic
loss aversion, Journal of Financial and Quantitative Analysis 44, 155-188.
Pastor, L., and R. Stambaugh, 2003, Liquidity risk and expected stock returns, Journal
of Political Economy 111, 642-685.
Patel, S., and S. Sarkissian, 2016, To group or not to group? Evidence from mutual fund
databases, Journal of Financial and Quantitative Analysis forthcoming.
32
Pesaran, M., 2006, Estimation and inference in large heterogeneous panels with a multi
factor error structure, Econometrica 74, 967-1012.
Pollet, J., and M. Wilson, 2008, How does size a�ect mutual fund behavior?, Journal of
Finance 63, 2941-2969.
Pool, V., N. Sto�man, and S. Yonker, 2012, No place like home: Familiarity in mutual
fund manager portfolio choice, Review of Financial Studies 25, 2563-2599.
Prather, L., and K. Middleton, 2002, Are n+1 heads better than one? The case of mutual
fund managers, Journal of Economic Behavior and Organization 47, 103-120.
Pronin, E., D. Y. Lin, and L. Ross, 2002, The bias blind spot: Perceptions of bias in self
versus others, Personality and Social Psychology Bulletin 28, 369-381.
Puetz, A., and S. Ruenzi, 2011, Overcon�dence among professional investors: Evidence
from mutual fund managers, Journal of Business Finance & Accounting 38, 684-712.
Sah, R., and J. Stiglitz, 1991, The quality of managers in centralized versus decentralized
organizations, Quarterly Journal of Economics 106, 289-295.
Sharpe, W., 1981, Decentralized investment management, Journal of Finance 36, 217-234.
Scheinkman, J., and W. Xiong, 2003, Overcon�dence and speculative bubbles, Journal of
Political Economy 111, 1183-1220.
Schrand, C., and S. Zechman, 2012, Executive overcon�dence and the slippery slope to
�nancial misreporting, Journal of Accounting and Economics 53, 311-329.
Seru, A., T. Shumway, and N. Sto�man, 2010, Learning by Trading, Review of Financial
Studies 23, 705-739.
Statman, M., S. Thorley, and K. Vorkink, 2006, Investor overcon�dence and trading
volume, Review of Financial Studies 19, 1531-1565.
Stoner, J., 1968, Risky and cautious shifts in group decisions, Journal of Experimental
Social Psychology 4, 442-459.
Sul, D., 2016, Spurious cross-sectional and mindless panel regressions, Working paper,
University of Texas, Dallas.
Sunstein, C., 2002, The Law of group polarization, Journal of Political Philosophy 10,
175-195.
Sunstein, C., and R. Hastie, 2014, Making dumb groups smarter, Harvard Business Review
33
12, 90-98.
Svenson, O., 1981, Are we all less risky and more skillful than our fellow drivers?, Acta
Psychologica 47, 143-148.
Tetlock, P., and G. Mitchell, 2009, Implicit Bias and Accountability Systems: What must
organizations do to prevent discrimination?, Research in Organization Behavior 29, 3-38.
Wallach, M., and H. Kogan., 1965, The Roles of information, discussion, and consensus
in group risk taking, Journal of Experimental Social Psychology 1, 1-19.
Weinstein, N., 1980, Unrealistic optimism about future life events, Journal of Personality
and Social Psychology, 39, 806-820.
West, R., R. Meserve, and K. Stanovich, 2012, Cognitive sophistication does not attenuate
the bias blind spot, Journal of Personality and Social Psychology 103, 506.
Winkler, J. and S. Taylor, 1979, Preference, expectations, and attributional bias: Two
�eld studies, Journal of Applied Social Psychology 9, 183-197.
34
A Appendix
A. Proof of Lemma 1
Following a good signal z = 1, manager M1 with beliefs given by De�nition 1 forms the
following posteriors belief regarding his skill:
P1{skilled|z = 1} =P1{skilled, z = 1}
P1{skilled, z = 1}+ P1{unskilled, z = 1}=
ppH + ∆
ppH + ∆ + (1− p)pL −∆(A.1)
Since the discretionary trading strategy yields a payo� of r = R with probability q if and
only if the manager is skilled, and pays zero otherwise, the expected payo� conditional on
observing z = 1 is:
E1{Rd|z = 1} = P1{skilled|z = 1}qR =ppH
ppH + (1− p)pLqR (A.2)
And the perceived variance of the payo� is:
V ar1{Rd|z = 1} = E1{R2d|z = 1}−E1{Rd|z = 1}2 = P1{skilled|z = 1}qR2−P1{skilled|z = 1}2q2R2
=ppH
ppH + (1− p)pLq
(1− ppH
ppH + (1− p)pLq
)R
2(A.3)
Similarly, M1's posterior belief regarding his skill after a bad signal z = 0 is:
P1{skilled|z = 0} =P1{skilled, z = 0}
P1{skilled, z = 0}+ P1{unskilled, z = 0}=
p(1− pH)
p(1− pH) + (1− p)(1− pL)(A.4)
Correspondingly, M1's perception of the expected value and variance of the discretionary
trading strategy Rd following a bad signal z = 0 are given by:
E1{Rd|z = 0} = P1{skilled|z = 0}qR =p(1− pH)
p(1− pH) + (1− p)(1− pL)qR (A.5)
V ar1{Rd|z = 0} =p(1− pH)
p(1− pH) + (1− p)(1− pL)q
(1− p(1− pH)
p(1− pH) + (1− p)(1− pL)q
)R
2
(A.6)
35
�
B. Proof of Proposition 1
Manager M1 solves the portfolio optimization problem (4) subject to beliefs given by
Lemma 1.
The �rst order condition of the optimization problem is:
w∗1|z =E1{Rd|z} −Rb
AV ar1{Rd|z}=
P1{Rd = 1|z}qR−Rb
AP1{Rd = 1|z}q(1− P1{Rd = 1|z}q)R2 (A.7)
We �rst establish the following result:
Lemma 2 Whenever the optimal weight on discretionary trading, w∗1|z, is positive, it is
increasing in P1{Rd = 1|z}.
Proof. Assume that the optimal weight is positive. Then P1{Rd = 1|z}qR > Rb.
Now note that:
δw∗1|zδP1{Rd = 1|z}
∝ Aq2R3P1{Rd = 1|z}(1− P1{Rd = 1|z}q) (A.8)
−Aq2R3P1{Rd = 1|z}(1− 2P1{Rd = 1|z}q) + AqR2Rb(1− 2P1{Rd = 1|z}q)
= Aq2R2P1{Rd = 1|z}q − AqR2
RbP1{Rd = 1|z}q + AqR2Rb(1− P1{Rd = 1|z}q)
≥ AqR2(P1{Rd = 1|z}qR−Rb),
which is greater than zero if P1{Rd = 1|z}qR > Rb. Hence, whenever the optimal weight on
the discretionary trading is positive, it is increasing in P1{Rd = 1|z}.
Now, let us compare {w∗1|z = 1} against {w∗1|z = 0}. Condition (5) establishes that
P1{Rd = 1|z = 1}qR > Rb, and therefore that {w∗1|z = 1} is strictly positive. If {w∗1|z = 0}
is zero, then we have {w∗1|z = 1} > {w∗1|z = 0} and we are done with the �rst part of
Proposition 1. If {w∗1|z = 0} is positive, then we can establish that {w∗1|z = 1} > {w∗1|z = 0}
using Lemma 2 in conjunction with the following result:
Lemma 3 P1{Rd = 1|z = 1} > P1{Rd = 1|z = 0}.
36
Proof. Recall from the proof of Proposition 1 that:
P1{Rd = 1|z = 1} =ppH + ∆
ppH + (1− p)pLq; P1{Rd = 1|z = 0} =
p(1− pH)
p(1− pH) + (1− p)(1− pL)(A.9)
So we have:
P1{Rd = 1|z = 1} ≤ ppHppH + (1− p)pL
q; P1{Rd = 1|z = 0} =p
p+ (1− p) pLpH
q (A.10)
Since pH > pL, we have p + (1− p) qLqH
< p, and hence the expression in (A.10) is greater
than pq.
Meanwhile, since 1− pL > 1− pH , we have:
P1{Rd = 1|z = 0} =p(1− pH)
p(1− pH) + (1− p)(1− pL)q =
p
p+ (1− p) 1−pL1−pH
q < pq (A.11)
Therefore, P1{Rd = 1|z = 1} is greater than P1{Rd = 1|z = 0}.
This completes the proof of the �rst part of Proposition 1 � that the overcon�dent manager
M1 engages in more discretionary trading following a good signal z = 1.
Now, in order to establish the second part of the proposition � that the discretionary
trading after a good signal is increasing in the manager's overcon�dence � we consider the
following comparative static:
δP1{Rd = 1|z = 1}δ∆
=1
ppH + (1− p)pL> 0 (A.12)
Thus, the manager's beliefs regarding the likelihood of a positive payo� following a good
signal, P1{Rd = 1|z = 1}, increase in the level of overcon�dence ∆. Recall that by condition
(5), the chosen weight on the discretionary strategy following a good signal is strictly positive.
Hence, we can again invoke Lemma 2, yielding that the manager's chosen weight on the
discretionary trading strategy likewise increases in the level of his overcon�dence ∆.
�
37
C. Proof of Proposition 2
First, recall from Lemma 1 that manager M1's beliefs regarding his own skill, conditional
on the realization of the signal z, are:
P1{Rd = 1|z = 1} =ppH + ∆
ppH + (1− p)pL; P1{Rd = 1|z = 0} =
p(1− pH)
p(1− pH) + (1− p)(1− pL)(A.13)
Using the same logic as in the proof of Lemma 1, we can also arrive at the beliefs that
other managers Mj 6=1 hold regarding M1's skill, conditional on the realization of the signal
z:
Pj 6=11{Rd = 1|z = 1} =ppH
ppH + (1− p)pL;Pj 6=1{Rd = 1|z = 0} =
p(1− pH)
p(1− pH) + (1− p)(1− pL)(A.14)
Note that: (1) all managers' beliefs are the same following a bad signal; and (2) P1{Rd =
1|z = 1} > Pj 6=1{Rd = 1|z = 1}.
Point (1) implies that all managers choose the same allocation into the proposed new
strategy following a bad signal. Hence, the weighted average of the allocation in the team-
managed fund is exactly the same as M1's choice in the single-managed fund. Thus, the
team-managed fund behaves identically to the single-managed fund following a bad signal.
Point (2) combined with Lemma 2 implies that the other managers Mj 6=1 choose lower
allocations to the new proposed strategy than the managerM1. Hence, the weighted average
in the team-managed fund is likewise lower than in the single-managed fund.
Since the two types of funds trade identically following a bad signal and the team-managed
fund trades less following a good signal, performance-induced trading (di�erence between
trading following good versus bad signals) is lower in team-managed funds.
�
38
0.7
0.8
0.9
1.0
Q1(L) Q2 Q3 Q4(H)
Fund
Tur
nove
r
Past Fund Performance Quartiles
Figure 1
Relation between past fund OARs and subsequent turnover. This �gure shows the relation betweenquartiles of past objective-adjusted gross fund returns (OARs) and subsequent turnover. The sample includesall domestic U.S. equity mutual funds (excluding index and sector funds) for the 1991-2015 period. Turnoveris the minimum of aggregated sales or aggregated purchases of securities during the year divided by theaverage 12-month total net assets of the fund. Fund performance is based on the annual objective-adjustedgross fund returns computed each year.
39
0.7
0.8
0.9
1.0
1.1
Q1(L) Q2 Q3 Q4(H)
Fund
Tur
nove
r
Fund Past Performance Quartiles
Single Team
(a) Single vs Team Funds
0.7
0.8
0.9
1.0
1.1
Q1(L) Q2 Q3 Q4(H)
Fund
Tur
nove
r
Fund Past Performance Quartiles
Single 2FM Team
(b) Single vs Two Manager Funds
0.7
0.8
0.9
1.0
1.1
Q1(L) Q2 Q3 Q4(H)
Fund
Tur
nove
r
Fund Past Performance Quartiles
Single 3FM Team
(c) Single vs Three Manager Funds
0.7
0.8
0.9
1.0
1.1
Q1(L) Q2 Q3 Q4(H)
Fund
Tur
nove
r
Fund Past Performance Quartiles
Single 4FM+ Team
(d) Single vs Four+ Manager Funds
Figure 2
Relation between past fund OARs and subsequent turnover for teams of di�erent sizes. This�gure shows the relation between quartiles of past objective-adjusted gross fund returns (OARs) and sub-sequent fund turnover for di�erent managerial team sizes. The sample includes all domestic U.S. equitymutual funds (excluding index and sector funds) for the 1991-2015 period. Turnover is the minimum ofaggregated sales or aggregated purchases of securities during the year divided by the average 12-month totalnet assets of the fund. The team-managed funds are divided into two manager funds (2FM), three managerfunds (3FM), and four or more manager funds (4FM+).
40
0.0
0.1
0.2
0.3
0.4
Q1(L) Q2 Q3 Q4(H)
Fund
Exc
ess
Turn
over
Fund Past Alpha Quartiles
Single Team
(a) Single vs Team Funds
0.0
0.1
0.2
0.3
0.4
Q1(L) Q2 Q3 Q4(H)
Fund
Exc
ess
Turn
over
Fund Alpha Quartiles
Single 2FM Team
(b) Single vs Two Manager Funds
0.0
0.1
0.2
0.3
0.4
Q1(L) Q2 Q3 Q4(H)
Fund
Exc
ess
Turn
over
Fund Past Alpha Quartiles
Single 3FM Team
(c) Single vs Three Manager Funds
0.0
0.1
0.2
0.3
0.4
Q1(L) Q2 Q3 Q4(H)
Fund
Exc
ess
Turn
over
Fund Past Alpha Quartiles
Single 4FM+ Team
(d) Single vs Four+ Manager Funds
Figure 3
Relation between past fund alpha and subsequent turnover. This �gure shows the relation betweenquartiles of past fund alphas based on the Fama-French �ve-factor model and subsequent fund excess turnoverfor di�erent managerial team sizes. The sample includes all domestic U.S. equity mutual funds (excludingindex and sector funds) for the 1991-2015 period. Excess Turnover (percent per year) for each fund is de�nedas the di�erence between the fund turnover in a given year and the median turnover for all funds with thesame fund investment objective in that year. The team-managed funds are divided into two manager funds(2FM), three manager funds (3FM), and four or more manager funds (4FM+).
41
Table
1
SummaryStatistics
PanelA:Distribution
ofsingle-managed
andteam
-managed
funds
Single
Team
Twomanagers
ThreeManagers
Four+
Managers
Year
Total
Number
Percent
Number
Percent
Number
Percent
Number
Percent
Number
Percent
1991
564
400
70.9
164
29.1
102
18.1
386.7
244.3
1992
630
424
67.3
206
32.7
115
18.3
579.0
345.4
1993
748
492
65.8
256
34.2
144
19.3
709.4
425.6
1994
875
542
61.9
333
38.1
194
22.2
849.6
556.3
1995
999
616
61.7
383
38.3
228
22.8
959.5
606.0
1996
1,121
666
59.4
455
40.6
262
23.4
119
10.6
746.6
1997
1,278
716
56.0
562
44.0
336
26.3
120
9.4
106
8.3
1998
1,470
805
54.8
665
45.2
385
26.2
162
11.0
118
8.0
1999
1,673
870
52.0
803
48.0
454
27.1
198
11.8
151
9.0
2000
1,780
875
49.2
905
50.8
494
27.8
231
13.0
180
10.1
2001
1,929
894
46.3
1035
53.7
554
28.7
244
12.6
237
12.3
2002
1,990
880
44.2
1110
55.8
584
29.3
274
13.8
252
12.7
2003
1,970
849
43.1
1121
56.9
588
29.8
282
14.3
251
12.7
2004
1,966
821
41.8
1145
58.2
587
29.9
286
14.5
272
13.8
2005
1,952
749
38.4
1203
61.6
591
30.3
278
14.2
334
17.1
2006
2,064
709
34.4
1355
65.6
604
29.3
290
14.1
461
22.3
2007
2,075
683
32.9
1392
67.1
613
29.5
299
14.4
480
23.1
2008
2,051
652
31.8
1399
68.2
613
29.9
298
14.5
488
23.8
2009
1,859
592
31.8
1267
68.2
542
29.2
286
15.4
439
23.6
2010
1,744
516
29.6
1228
70.4
547
31.4
303
17.4
378
21.7
2011
1,656
477
28.8
1179
71.2
518
31.3
311
18.8
350
21.1
2012
1,554
455
29.3
1099
70.7
471
30.3
284
18.3
344
22.1
2013
1,465
401
27.4
1064
72.6
432
29.5
280
19.1
352
24.0
2014
1,429
403
28.2
1026
71.8
416
29.1
281
19.7
329
23.0
2015
1,386
397
28.6
989
71.4
416
30.0
251
18.1
322
23.2
Total
38,228
15,884
22,344
10,790
5,421
6,133
42
Panel B: Characteristics of single-managed and team-managed funds (Full Sample 1991 - 2015)
Single-Managed Funds Team-Managed FundsObs Mean SD Obs Mean SD Di� (Team-Single)
Turnover 14,748 0.8615 0.8155 21,464 0.7846 0.6507 -0.0769***Excess Turnover 14,465 0.2284 0.8219 21,235 0.1557 0.6454 -0.0728***Fund Size (millions) 15,774 1,143 4,284 22,276 1,295 5,405 151.5***Fund Age (years) 15,883 13.22 14.09 22,343 13.51 13.40 0.29**Family Size (billions) 15,665 39.14 109.06 22,243 20.45 55.95 -18.69***Fund Fees 15,383 0.0130 0.0051 21,878 0.0123 0.0043 -0.0007***Fund Flows 15,638 0.3689 1.5364 22,160 0.3021 1.4554 -0.0668***OAR (%/m) 15,884 0.0109 0.1309 22,344 0.0083 0.1126 -0.0026**α(FF5) (%/m) 13,595 0.0759 0.7229 19,757 0.0613 0.4282 -0.0145**Return Volatility 15,830 0.0476 0.0554 22,301 0.0471 0.0545 -0.0005Mgr Industry Tenure 15,252 16.4833 9.5375 21,490 16.6394 7.0002 -0.1561*Female (%) 15,306 0.0921 0.2800 22,151 0.0913 0.1842 -0.0008
This table shows the annual distribution of team-managed and team-managed funds (Panel A) and thesummary statistics of their fund and managerial characteristics (Panel B). The sample includes all domesticU.S. equity mutual funds (excluding index and sector funds) for the 1991-2015 period. Turnover is theminimum of aggregated sales or aggregated purchases of securities during the year divided by the average12-month total net assets of the fund. Excess Turnover (percent per year) of each fund is de�ned as thedi�erence between the fund turnover in a given year and the median turnover for all funds with the same fundinvestment objective in that year. Fund Size (millions of dollars) is the total net assets under managementof a fund in a given year. Fund Age (years) is the di�erence between the fund's inception year and thecurrent year. Family Size (billions of dollars) is measured by the total net assets under management of thefund complex to which the fund belongs at the end of the calendar year. Fund Fees (percent per year) isthe annual total expense ratio of the fund. Return Volatility (percent per year) is the standard deviationof monthly gross fund returns over the past 12 months. Flows is de�ned as the net growth in the total netassets of a fund, as a percentage of the fund's total net assets, adjusted for prior the year return. ExcessTurnover, Fund Fees, turnover, and fund �ows are winsorized at 1% and 99% levels. OAR is the annualobjective-adjusted gross fund returns computed each year. α(FF5) is the alpha based on Fama and French(2015) �ve-factor model which adds pro�tability (RMW) and investment (CMA) factors to the Fama andFrench (1993) three-factor model. Manager Industry Tenure (years) is the number of years the fund managerhas been within the fund industry. Female is de�ned as a proportion of female managers in a fund. Di�(Team-Single) is the di�erence between team-managed and single-managed funds. ***, **, and * denotesigni�cance at the 1%, 5%, and 10% levels, respectively.
43
Table
2
TopPerformanceandTurnoverRelation
Top
25th
PercentilePerform
ance
Top
20th
PercentilePerform
ance
Top
10th
PercentilePerform
ance
DV:Turnover
ex i,t
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
Team
i,t−
1×
Perf i,t−1
-0.0923***
-0.0897***
-0.0863***
-0.0795**
-0.0730*
-0.0700**
-0.0858**
-0.1014*
-0.1159**
(0.005)
(0.008)
(0.006)
(0.026)
(0.053)
(0.041)
(0.018)
(0.076)
(0.034)
Perf i,t−1
0.1381***
0.0938***
0.0905***
0.1412***
0.0882**
0.0824***
0.1826***
0.1561***
0.1461***
(0.001)
(0.009)
(0.005)
(0.000)
(0.013)
(0.007)
(0.000)
(0.007)
(0.005)
Team
i,t−
1-0.0389*
-0.0227
-0.0399*
-0.0465**
-0.0307
-0.0473**
-0.0497**
-0.0349*
-0.0489**
(0.095)
(0.307)
(0.055)
(0.047)
(0.170)
(0.023)
(0.023)
(0.096)
(0.014)
Flows i,t−1
-0.0172*
-0.0143*
-0.0175**
-0.0145*
-0.0190**
-0.0155**
(0.055)
(0.064)
(0.049)
(0.060)
(0.035)
(0.045)
FF5 i
,t−1
-0.0155
-0.0112
-0.0150
-0.0101
-0.0203
-0.0133
(0.566)
(0.653)
(0.581)
(0.689)
(0.460)
(0.590)
Size i
,t−1
-0.0587***
-0.0546***
-0.0587***
-0.0546***
-0.0588***
-0.0545***
(0.000)
(0.000)
(0.000)
(0.000)
(0.000)
(0.000)
Volatilityi,t−
10.7566
0.6190
0.7566
0.6202
0.7448
0.6125
(0.291)
(0.309)
(0.289)
(0.307)
(0.294)
(0.311)
Expenses i,t−1
0.2082***
0.2109***
0.2080***
0.2110***
0.2053***
0.2089***
(0.000)
(0.000)
(0.000)
(0.000)
(0.000)
(0.000)
Age
i,t−
10.0009
0.0177
0.0007
0.0174
0.0011
0.0176
(0.950)
(0.210)
(0.958)
(0.217)
(0.935)
(0.212)
Fam
ilySize i
,t−1
0.0342***
0.0221**
0.0343***
0.0220*
0.0345***
0.0215*
(0.000)
(0.049)
(0.000)
(0.050)
(0.000)
(0.056)
FC
i0.0827***
0.0330
0.0827***
0.0329
0.0831***
0.0333
(0.000)
(0.357)
(0.000)
(0.359)
(0.000)
(0.353)
Tenure
i,t−
1-0.0106***
-0.0079***
-0.0106***
-0.0078***
-0.0106***
-0.0079***
(0.000)
(0.000)
(0.000)
(0.000)
(0.000)
(0.000)
Fem
ale i
,t−1
-0.0017
-0.0284
-0.0008
-0.0277
-0.0002
-0.0275
(0.970)
(0.514)
(0.986)
(0.526)
(0.997)
(0.529)
ObjectiveFE
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Fam
ilyFE
No
No
Yes
No
No
Yes
No
No
Yes
Adj.R
20.026
0.083
0.184
0.026
0.083
0.183
0.029
0.084
0.184
Obs.
33,182
26,064
26,063
33,182
26,064
26,063
33,182
26,064
26,063
Di�
(Team-Single)
-0.1312***
-0.1124***
-0.1261***
-0.1260***
-0.1037***
-0.1173***
-0.1354***
-0.1364**
-0.1648***
p-value
(0.000)
(0.001)
(0.000)
(0.000)
(0.006)
(0.004)
(0.000)
(0.016)
(0.003)
44
Table 2 �(continued from previous page)
This table shows the e�ect of fund performance on subsequent excess fund turnover. The sample includes alldomestic U.S. equity mutual funds (excluding index and sector funds) for the 1991-2015 period. Three topperformance ranges, Perf, are considered: top 25% (�rst three columns), top 20% (second three columns),and top 10% (last three columns). Team is de�ned as a dummy variable that equals one if the fund hastwo or more fund managers and zero if the fund has only one fund manager at the end of the precedingcalendar year. FC is the indicator variable for a �nancial center, which is equal to one if a fund is locatedin the following six cities: Boston, Chicago, Los Angeles, Philadelphia, New York, and San Francisco. Allother characteristics are de�ned in Table 1. Intercept is included in each regression but its estimates are notreported. Fixed e�ects include fund investment objective and fund family �xed e�ects. Standard errors areclustered by fund and year. The last two rows show the di�erence in excess turnover between team-managedand single-managed funds, Di� (Team-Single), as well as the p-value of the corresponding F-test. ***, **,and * denote signi�cance at the 1%, 5%, and 10% levels, respectively.
45
Table
3
TopPerformanceandTurnoverRelationAcross
ManagerialTeamsofDi�erentSizes
Singlevs2FM
Team
Singlevs3FM
Team
Singlevs4+
FM
Team
DV:Turnover
ex i,t
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
Team
i,t−
1×
Perf i,t−1
-0.0633*
-0.0712**
-0.0674**
-0.0914*
-0.0684
-0.1242***
-0.1373***
-0.1304***
-0.1201***
(0.080)
(0.047)
(0.043)
(0.060)
(0.176)
(0.002)
(0.000)
(0.005)
(0.005)
Perf i,t−1
0.1384***
0.0915**
0.0854***
0.1398***
0.0878**
0.0852***
0.1374***
0.0889**
0.0874***
(0.001)
(0.010)
(0.006)
(0.001)
(0.013)
(0.009)
(0.001)
(0.013)
(0.007)
Team
i,t−
1-0.0433
-0.0421*
-0.0515**
-0.0156
-0.0002
-0.0090
-0.0509*
-0.0086
-0.0114
(0.110)
(0.094)
(0.025)
(0.597)
(0.995)
(0.750)
(0.050)
(0.738)
(0.669)
Controls
No
Yes
Yes
No
Yes
Yes
No
Yes
Yes
ObjectiveFE
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Fam
ilyFE
No
No
Yes
No
No
Yes
No
No
Yes
Adj.R
20.026
0.090
0.202
0.025
0.083
0.207
0.030
0.088
0.191
Obs.
22,251
17,357
17,356
18,176
14,105
14,104
19,657
14,948
14,947
Di�
(Team-Single)
-0.1066***
-0.1133***
-0.1261***
-0.1071**
-0.0686
-0.1332***
-0.1882***
-0.1390***
-0.1315***
p-value
(0.001)
(0.001)
(0.000)
(0.017)
(0.146)
(0.002)
(0.000)
(0.005)
(0.008)
Thistableshow
sthee�ectoffundperform
ance
onsubsequentexcess
turnover
across
fundmanager
teamsofdi�erentsizes.
Thesampleincludes
all
domesticU.S.equitymutualfunds(excludingindex
andsectorfunds)
forthe1991-2015period.Perfisanindicatorforthefund'sperform
ance
inyeart−
1fallingin
thetopquartile.
Team
isde�ned
asadummyvariablethatequalsoneifthefundhastwoormore
fundmanagersandzero
ifthe
fundhasonly
onefundmanager
attheendofthecalendaryear.
Controlsare
thesamefundandmanager
characteristics
asin
Table2.Interceptis
included
ineach
regressionbutitsestimatesare
notreported.Fixed
e�ectsincludefundinvestm
entobjectiveandfundfamily�xed
e�ects.Standard
errors
are
clustered
byfundandyear.
Thelast
tworowsshow
thedi�erence
inexcess
turnover
betweenteam-m
anaged
andsingle-m
anaged
funds,
Di�
(Team-Single),aswellasthep-valueofthecorrespondingF-test.***,**,and*denote
signi�cance
atthe1%,5%,and10%
levels,respectively.
46
Table 4
Changes in Managerial Structure and Turnover
Single to Team Team to SingleDV: ∆Turnoverexi,t (1) (2) (3) (4) (5) (6)
∆MSi,t−1 × Perfi,t−1 -0.1495*** -0.0929** -0.0598 -0.0257 0.0562 0.0521(0.002) (0.048) (0.140) (0.796) (0.598) (0.513)
Perfi,t−1 0.1325*** 0.1095*** 0.0676** 0.0241 0.0535 0.0148(0.007) (0.005) (0.045) (0.779) (0.650) (0.861)
∆MSi,t−1 0.0208 -0.0074 -0.0234 0.0281 0.0113 0.0003(0.521) (0.830) (0.467) (0.709) (0.877) (0.996)
Controls No Yes Yes No Yes Yes
Objective FE Yes Yes Yes Yes Yes YesFamily FE No No Yes No No YesAdj. R2 0.037 0.138 0.333 0.013 0.062 0.456Obs. 8,097 6,709 6,709 2,362 1,809 1,809
Di� (Team-Single) -0.1286*** -0.1001** -0.0832* 0.0024 0.0676 0.0524p-value (0.002) (0.032) (0.068) (0.982) (0.581) (0.576)
This table shows the e�ect of changes in managerial structure of funds on subsequent excess turnover. Thesample includes all domestic U.S. equity mutual funds (excluding index and sector funds) for the 1991-2015period. Perf is an indicator variable for the fund's performance in year t− 1 falling within the top quartile.∆MSi,t−1 is the change in managerial structure of fund i at time t− 1. For columns 1-3 ∆MS is a dummythat takes the value of one if a fund is team-managed at time t-1 but was single-managed at time t-2.For columns 4-6 ∆MS is a dummy that takes the value of one if a fund is single-managed at time t − 1but was team-managed at time t − 2. A fund is team-managed if it has two or more fund managers atthe end of the preceding calendar year. The �rst three columns report the results for funds that becometeam-managed at some point in the sample period. The last three columns report the results for funds thatbecome single-managed at some point in the sample period. There are 233 funds that switch from single-to team-management and 162 funds that switched from team- to single-management. Controls are the samefund and manager characteristics as in Table 2. Intercept is included in each regression but its estimatesare not reported. Fixed e�ects include fund investment objective and fund family �xed e�ects. Standarderrors are clustered by fund and year. The last two rows show the di�erence in excess turnover betweenteam-managed and single-managed funds, Di� (Team-Single), as well as the p-value of the correspondingF-test. ***, **, and * denote signi�cance at the 1%, 5%, and 10% levels, respectively.
47
Table 5
Matching Tests on Changes in Managerial Structure and Turnover
Panel A: Nearest neighbor (one-to-one matches)
Single to Team Team to Single
DV: ∆Turnoverexi,t (1) (2) (3) (4) (5) (6)
Treated Funds -0.1492** -0.1534*** -0.2105*** 0.0535 0.0558 0.0657
(0.014) (0.006) (0.000) (0.166) (0.133) (0.462)
Controls No Yes Yes No Yes Yes
Objective FE Yes Yes Yes Yes Yes Yes
Family FE No No Yes No No Yes
Adj. R2 0.024 0.099 0.325 0.028 0.037 0.143
Obs. 466 464 464 324 324 324
Panel B: Nearest �ve neighbors (one-to-�ve matches)
Single to Team Team to Single
DV: ∆Turnoverexi,t (1) (2) (3) (4) (5) (6)
Treated Funds -0.0948** -0.0875** -0.1156*** 0.0698* 0.0594* 0.0648
(0.019) (0.023) (0.001) (0.056) (0.088) (0.155)
Controls No Yes Yes No Yes Yes
Objective FE Yes Yes Yes Yes Yes Yes
Family FE No No Yes No No Yes
Adj. R2 0.011 0.072 0.172 0.012 0.035 0.206
Obs. 792 790 790 553 553 553
This table shows the e�ect of changes in managerial structure of funds on subsequent excess turnover usingthe treated and matched fund samples. The sample includes all domestic U.S. equity mutual funds (excludingindex and sector funds) for the 1991-2015 period. The �rst three columns report the results for funds thatbecome team-managed at time t-1. The last three columns report the results for funds that become single-managed at time t-1. There are 233 funds that switch from single- to team-management and 162 funds thatswitch from team- to single-management. To create the control sample, we use propensity score matchingapproach using logistic regressions to identify funds that share similar observable fund characteristics withthe treatment group. Each fund that switches its managerial structure is matched, with replacement, to thefund with the closest propensity score based on fund characteristics such as performance, size, age, �ows,expenses, family size, investment objective, as well as manager characteristics such as industry tenure andgender composition over the same period. We then calculate the di�erence in excess turnover of funds thatswitched from single (team) to team (single) management in the next period and funds that are in thematched sample. Panel A shows the estimation results for one-to-one fund matching, while Panel B displaysthe results for one-to-�ve fund matching. Team is de�ned as a dummy variable that equals one if the fundhas two or more fund managers and zero if the fund has only one fund manager at the end of the calendaryear. Controls are the same fund and manager characteristics as in Table 2. Intercept is included in eachregression but its estimates are not reported. Fixed e�ects include fund investment objective and fund family�xed e�ects. Standard errors are clustered by fund and year. ***, **, and * denote signi�cance at the 1%,5%, and 10% levels, respectively.
48
Table 6
Placebo Tests on Changes in Managerial Structure
Panel A: Between 25 to 50 Percentile (Quartile 2)
Single to Team Team to Single
DV: ∆Turnoverexi,t (1) (2) (3) (4) (5) (6)
Treated Funds -0.0032 -0.0086 -0.0359 0.0081 0.0101 0.0233
(0.944) (0.852) (0.602) (0.893) (0.860) (0.771)
Controls No Yes Yes No Yes Yes
Objective FE Yes Yes Yes Yes Yes Yes
Family FE No No Yes No No Yes
Adj. R2 0.001 -0.006 0.017 0.024 0.038 0.364
Obs. 618 617 617 335 334 334
Panel B: Between 50 to 75 Percentile (Quartile 3)
Single to Team Team to Single
DV: ∆Turnoverexi,t (1) (2) (3) (4) (5) (6)
Treated Funds 0.0108 0.0127 0.0244 0.0028 0.0180 0.1160*
(0.776) (0.742) (0.550) (0.962) (0.753) (0.072)
Controls No Yes Yes No Yes Yes
Objective FE Yes Yes Yes Yes Yes Yes
Family FE No No Yes No No Yes
Adj. R2 0.001 0.007 0.145 0.021 0.014 0.361
Obs. 576 573 573 340 338 338
This table shows the e�ect of changes in managerial structure of funds on subsequent excess turnover usingthe placebo tests on treated and matched fund samples. The sample includes all domestic U.S. equity mutualfunds (excluding index and sector funds) for the 1991-2015 period. Panel A shows the results for the secondpast performance quartile; Panel B - for the third quartile. The performance metric is based on the Fama-French �ve-factor alpha. The �rst three columns report the results for funds that become team-managedat time t-1. The last three columns report the results for funds that become single-managed at time t-1.Team is de�ned as a dummy variable that equals one if the fund has two or more fund managers and zero ifthe fund has only one fund manager at the end of the preceding calendar year. Controls are the same fundand manager characteristics as in Table 2. Intercept is included in each regression but its estimates are notreported. Fixed e�ects include fund investment objective and fund family �xed e�ects. Standard errors areclustered by fund and year. ***, **, and * denote signi�cance at the 1%, 5%, and 10% levels, respectively.
49
Table 7
Fama-Macbeth Regressions: Cross-Sectional Tests
DV: Turnoverexi,t (1) (2) (3)
Teami,t−1 × Perfi,t−1 -0.0492** -0.0764*** -0.0664**
(0.017) (0.002) (0.038)
Perfi,t−1 0.0908** 0.1312*** 0.1083***
(0.039) (0.000) (0.000)
Teami,t−1 -0.0533*** -0.0477*** -0.0556***
(0.000) (0.001) (0.000)
Controls Yes Yes Yes
Objective FE Yes Yes Yes
Family FE No No Yes
Adj. R2 0.035 0.138 0.625
Obs. 35,700 28,195 28,195
Di� (Team - Single) -0.1025*** -0.1225*** -0.1220***
P-value (F-test) (0.001) (0.000) (0.003)
This table shows the e�ect of fund performance on subsequent excess turnover using the Fama-MacBethcross-sectional regression method. The sample includes all domestic U.S. equity mutual funds (excludingindex and sector funds) for the 1991-2015 period. Perf is a dummy variable for past performance in the topquartile. Team is de�ned as a dummy variable that equals one if the fund has two or more fund managersand zero if the fund has only one fund manager at the end of the preceding calendar year. Controls arethe same fund and manager characteristics as in Table 2. Intercept is included in each regression but itsestimates are not reported. Fixed e�ects include fund investment objective and fund family �xed e�ects.The last two rows show the di�erence in excess turnover between team-managed and single-managed funds,Di� (Team-Single), as well as the p-value of the corresponding F-test. ***, **, and * denote signi�cance atthe 1%, 5%, and 10% levels, respectively.
50
Table 8
Alternative Explanations
Industry Experience Net Fund FlowsDV: Turnoverexi,t <10 years ≥10 years In�ows Out�ows Tournaments Males Only
Teami,t−1 × Top Perfi,t−1 -0.1061* -0.0693** -0.0536 -0.1349*** -0.0878*** -0.0854***(0.051) (0.049) (0.151) (0.000) (0.005) (0.009)
Top Perfi,t−1 0.1225*** 0.0660** 0.0751** 0.1445*** 0.0936*** 0.0925***(0.008) (0.048) (0.042) (0.000) (0.004) (0.005)
Teami,t−1 -0.1009*** -0.0200 -0.0384 -0.0502* -0.0402* -0.0331(0.001) (0.382) (0.124) (0.057) (0.053) (0.130)
Flowsi,t−1 -0.0058 -0.0169** 0.0016 -0.2594*** -0.0148* -0.0150*(0.674) (0.048) (0.778) (0.000) (0.059) (0.067)
FF5i,t−1 -0.0651 -0.0135 0.0215 -0.0577 -0.0103 -0.0093(0.163) (0.603) (0.266) (0.105) (0.662) (0.698)
Sizei,t−1 -0.0688*** -0.0522*** -0.0524*** -0.0467*** -0.0547*** -0.0604***(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Volatilityi,t−1 4.2573*** 0.3443 0.3424 2.5360*** 0.4969(0.000) (0.389) (0.359) (0.000) (0.349)
Expensesi,t−1 0.1901*** 0.2139*** 0.1729*** 0.2204*** 0.2132*** 0.2087***(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Agei,t−1 0.0168 0.0186 0.0061 0.0094 0.0170 0.0073(0.423) (0.223) (0.695) (0.583) (0.234) (0.664)
Family Sizei,t−1 0.0299* 0.0187 0.0267** 0.0147 0.0220** 0.0194(0.094) (0.108) (0.046) (0.262) (0.050) (0.140)
FCi 0.1166*** 0.0031 0.0403 0.0254 0.0321 0.0465(0.009) (0.940) (0.272) (0.559) (0.371) (0.231)
Tenurei,t−1 0.0015 -0.0063*** -0.0094*** -0.0070*** -0.0079*** -0.0067***(0.810) (0.000) (0.000) (0.000) (0.000) (0.000)
Femalei,t−1 -0.0646 -0.0271 0.0149 -0.0613 -0.0278(0.307) (0.575) (0.789) (0.157) (0.529)
AbsVoli,t 0.0800(0.667)
Objective FE Yes Yes Yes Yes Yes YesFamily FE Yes Yes Yes Yes Yes YesAdj. R2 0.261 0.188 0.189 0.197 0.182 0.195Obs. 5,176 20,887 11,421 14,642 26,063 20,785
Di� (Team-Single) -0.2069*** -0.0893** -0.0919** -0.1851*** -0.1280*** -0.1185***p-value (0.001) (0.012) (0.024) (0.004) (0.000) (0.000)
This table shows the e�ect of fund performance on subsequent excess fund turnover using alternative expla-nation settings. The �rst two columns split the sample based on manager industry experience. The third andfourth columns split the sample based on net fund in�ows and net fund out�ows. The �fth column estimationreplaces fund return volatility in year t with the absolute value of the changes in fund return volatility fromyear t-1 to t, AbsVoli,t. The last column uses a subsample of funds managed by male managers, both single-managed and team-managed. The other control variables are the same fund and manager characteristics asin Table 2. Fixed e�ects include fund investment objective and fund family �xed e�ects. Standard errors areclustered by fund and year. The last two rows show the di�erence in excess turnover between team-managedand single-managed funds, Di� (Team-Single), as well as the p-value of the corresponding F-test. ***, **,and * denote signi�cance at the 1%, 5%, and 10% levels, respectively.
51
Table
9
Robustness:AlternativePerformanceMeasuresandSpeci�cations
OAR
α(C4)
α(PS)
α(FF5)
DV:Turnover
ex i,t
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
F-test:Di�
(Team-Single)
-0.0657**
-0.0872***
-0.0658**
-0.0696**
-0.0770**
-0.0814**
-0.0679***
-0.1347***
P-value
(0.043)
(0.005)
(0.050)
(0.037)
(0.025)
(0.016)
(0.000)
(0.000)
Controls
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
ObjectiveFE
Yes
Yes
Yes
Yes
Yes
Yes
Yes
No
Fam
ilyFE
No
Yes
No
Yes
No
Yes
No
Yes
Cluster
(Fund,Time)
Yes
Yes
Yes
Yes
Yes
Yes
No
Yes
FundFE
No
No
No
No
No
No
Yes
No
Objective×
TimeFE
No
No
No
No
No
No
No
Yes
Obs.
29,512
29,508
26,064
26,063
26,064
26,063
26,064
26,064
Thistableshow
sthee�ectoffundperform
ance
onsubsequentexcess
fundturnover
usingalternative
speci�cations.
Itonly
reportsthedi�erence
inexcessturnover
betweenteam-m
anaged
andsingle-m
anaged
funds,Di�
(Team-Single)aswellasthep-valueofthecorrespondingF-test.Thesample
includes
alldomesticU.S.equitymutualfunds(excludingindex
andsectorfunds)
forthe1991-2015period.Perfisanindicatorvariable
forpast
perform
ance
inthetopquartile.
OARistheobjective-adjusted
return;α
(C4)isthealphabasedontheCarhart
(1997)four-factormodel;α
(PS
5)
isthesimilarlycomputedrisk-adjusted
return
from
the�ve-factormodel,whichaddstheliquidityfactorofPastorandStambaugh(2003)to
the
Carhart
(1997)model.α
(FF
5)istheFama-French
�ve-factoralpha,asde�ned
inTable1.Team
isde�ned
asadummyvariablethatequalsoneif
thefundhastwoormore
fundmanagersandzero
ifthefundhasonly
onefundmanager
attheendoftheprecedingcalendaryear.
Controlsare
thesamefundandmanager
characteristics
asin
Table2.Interceptisincluded
ineach
regressionbutitsestimatesare
notreported.Fixed
e�ects
includefundinvestmentobjectiveandfundfamily�xed
e�ects.Speci�cation(7)replacesfundfamily�xed
e�ects
withindividualfund�xed
e�ects.
Speci�cation(8)replace
investmentobjective�xed
e�ects
withinvestmentobjectivetimes
year�xed
e�ects.Standard
errors
are
clustered
byfund
andyear.***,**,and*denote
signi�cance
atthe1%,5%,and10%
levels,respectively.
52
Table 10
Overcon�dence Induced Trading and Future Fund Performance
Panel A: Top Quartile Past Performance
Single-Managed Funds Team-Managed FundsDV: Fund Returns OARt α(C4)t α(PS5)t α(FF5)t OARt α(C4)t α(PS5)t α(FF5)t
Turnoverexi,t−1 0.0083 -0.0287 -0.0827 -0.0519 0.0025 -0.0227 -0.0449 0.0003(0.394) (0.572) (0.207) (0.171) (0.578) (0.567) (0.371) (0.992)
Controls Yes Yes Yes Yes Yes Yes Yes Yes
Objective × Time FE Yes Yes Yes Yes Yes Yes Yes YesFamily FE Yes Yes Yes Yes Yes Yes Yes YesAdj. R2 0.733 0.130 0.062 0.075 0.834 0.170 0.166 0.106Obs. 3,135 2,997 2,994 3,021 4,620 4,612 4,611 4,384
Panel B: Top Decile Past Performance
Single-Managed Funds Team-Managed FundsDV: Fund Returns OARt α(C4)t α(PS5)t α(FF5)t OARt α(C4)t α(PS5)t α(FF5)t
Turnoverexi,t−1 0.0154 -0.1016 -0.1169*** -0.0889 -0.0033 -0.0661 -0.0828 -0.0897(0.280) (0.194) (0.001) (0.105) (0.593) (0.303) (0.314) (0.216)
Controls Yes Yes Yes Yes Yes Yes Yes Yes
Objective × Time FE Yes Yes Yes Yes Yes Yes Yes YesFamily FE Yes Yes Yes Yes Yes Yes Yes YesAdj. R2 0.691 0.112 0.135 0.050 0.838 0.168 0.186 0.114Obs. 1,154 1,156 1,212 1,189 1,719 1,757 1,728 1,637
This table shows the relation between excess turnover and future fund performance, conditional on top pastperformance. The sample includes all domestic U.S. equity mutual funds (excluding index and sector funds)for the 1991-2015 period. OAR are objective-adjusted returns, α(C4) is the Carhart alpha, α(PS5) is thePastor-Stambaugh alpha, and α(FF5) is the Fama-French �ve-factor alpha, as de�ned in Table 1. ExcessTurnover (percent per year) of each fund is de�ned as the di�erence between the fund's turnover in a givenyear and the median turnover for all funds with the same fund investment objective in that year. Controlsare the same fund and manager characteristics as in Table 2. Intercept is included in each regression butits estimates are not reported. Fixed e�ects include fund investment objective times year �xed e�ects, aswell as and fund family �xed e�ects. Standard errors are clustered by fund and year. ***, **, and * denotesigni�cance at the 1%, 5%, and 10% levels, respectively.
53