Wulf Meyer 2016

Journal of Behavioral Finance
ISSN: 1542-7560 (Print) 1542-7579 (Online) Journal homepage: http://www.tandfonline.com/loi/hbhf20
Irrational Mutual Fund Managers: Explaining

Differences in Their Behavior
Sina Wulfmeyer
To cite this article: Sina Wulfmeyer (2016) Irrational Mutual Fund Managers: Explaining
Differences in Their Behavior, Journal of Behavioral Finance, 17:2, 99-123, DOI:
10.1080/15427560.2016.1133621
To link to this article: http://dx.doi.org/10.1080/15427560.2016.1133621
Published online: 25 May 2016.
Submit your article to this journal
Article views: 9
View related articles
View Crossmark data
Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=hbhf20
Download by: [University of Cambridge]
Date: 02 June 2016, At: 02:32
JOURNAL OF BEHAVIORAL FINANCE

2016, VOL. 17, NO. 2, 99123
http://dx.doi.org/10.1080/15427560.2016.1133621
Irrational Mutual Fund Managers: Explaining Differences in Their Behavior

Sina Wulfmeyer
Downloaded by [University of Cambridge] at 02:32 02 June 2016
University of St. Gallen
ABSTRACT
KEYWORDS
Documenting the disposition effect for a large sample of mutual fund managers in the United
States, we nd that stock-level characteristics explain the cross-sectional variation of the effect. The
disposition effect, which is the tendency to sell winner stocks too early and hold on to loser stocks
for too long, is more pronounced for fund managers who invest in stocks that are more difcult to
value. Using different measures of stock and market uncertainty, we show that mutual fund
managers display a stronger disposition-driven behavior when stocks are more difcult to value. We
also nd that the level of the disposition effect is monotonically increasing with the level of
systematic risk (i.e., beta). In addition, we document that the trading behavior of mutual fund
managers is partly driven by attention-grabbing stocks (dividend-paying stocks). Overall, our results
suggest that stock-level uncertainty and trading of attention-grabbing stocks amplify the
disposition effect and that differences in the effect can be explained by mutual fund managers
investment styles. Given that mutual funds hold a large fraction of the U.S. equity market, our
ndings add to the ongoing discussion whether professional investors can create stock mispricings
and shed new light on market efciency.
Mutual funds; Disposition

effect; Stock characteristics;
Portfolio management; Stock
valuation
Introduction
More than $11.6 trillion are invested in equity mutual
funds in 20121, and despite the growing number of index
funds and exchange-traded funds (ETFs), the vast majority of funds remains actively managed. However,
research has consistently shown that on average actively
managed portfolios do not outperform their benchmarks2. Lately, the presence of the disposition effect
among mutual fund managers has been put forward as a
major behavioral-based cause for underperformance
(e.g., Singal and Xu [2011], Scherbina and Jin [2011],
Cici [2010]). The disposition effect refers to the tendency
of investors to prefer holding on to loser stocks rather
than selling their winner stocks. It is based on key features of mental accounting (Thaler [1985]) and prospect
theory (Kahneman and Tversky [1979]). Our focus is on
the analysis of the disposition effect and how stockrelated characteristics inuence the magnitude of the
documented behavioral bias.
The goal of this paper is twofold. First, we analyze to
what extend the disposition effect is present among U.S.
equity mutual fund managers and whether the effect is
more prominent among certain stock holdings. We shed
new light on the question whether professional investors
behave rationally or whether they show signs of irrationality. Specically, we examine under which circumstances the disposition effect is evident among a sample of
actively managed, U.S. mutual fund managers from 1980
to 2010. While extensive research exists on the behavior
of individual investors, there is rather little and mixed
evidence on the effect of biased behavior among mutual
fund managers. It is questionable why the disposition
effect persists since merely pointing out any sort of bias
should cause investors to adjust their behavior and to
make them disappear.
Second, we test the hypothesis whether the cross-sectional variation in the magnitude of the disposition effect
can be explained by differences in investment styles. Previous studies focus on the stock-picking abilities of topperforming mutual fund managers: positive alphas suggest that the respective fund manager has skills (e.g.,
Wermers [2000], Kacperczyk, Sialm, and Zheng [2008]).
However, specic characteristics of the stock holdings of
fund managers and why some managers hold top-performing while other hold underperforming stocks have
not been fully explored yet. We look at the relationship
between the disposition effect and various stock characteristics. Furthermore, we analyze the relationship
CONTACT Sina Wulfmeyer

sina.wulfmeyer@gmail.com
Chair of Finance Prof. Ammann, Swiss Institute for Banking and Finance, University of St. Gallen,
Rosenbergstrasse 52, 9000 St. Gallen, Switzerland.
We used STATA to perform the analyses in the empirical section.
2016 The Institute of Behavioral Finance
100
S. WULFMEYER
between the magnitude of the stock-level disposition

effect and valuation uncertainty. Experimental evidence
indicates that people are more likely to use heuristics
and display a higher level of behavioral biases, when they
are faced with more difcult situations (e.g., Kahneman
and Tversky [1979], Kahneman [2003]). Theoretical
nancial models (Daniel, Hirshleifer, and Subrahmanyam [1998], Hirshleifer [2001]) formalize the experimental results of decision making in an investment
environment and state that investors biases will be
stronger when stocks are harder to value. We also test
the idea that there is ight to quality when risk or market volatility increases. The ight to quality hypothesis
states that when market uncertainty raises, investors
reduce their stock holdings, and ee to safer assets such
as bonds (Caballero and Krishnamurthy (2008)). This
hypothesis implies that we expect the level of the disposition effect to decrease when market risk raises. We analyze how the disposition bias is affected when mutual
fund managers are faced with increased valuation uncertainty and market risk.
Using a comprehensive sample of holdings of actively
managed, U.S. equity mutual funds from 1980 to 2010,
our analysis delivers two main results. First, we provide
evidence that mutual fund managers are prone to the
disposition effect. Using Odeans [1998] methodology of
calculating the disposition effect, we nd that mutual
fund managers show on average a disposition effect of
3.96%. Our ndings suggest that not only individual
investors but also professional investors are likely to
make biased decisions in their investment process. Our
results are signicant and more pronounced for smaller
and younger funds as well as for funds with a higher fee
structure.
Second, we show that the level of the disposition effect
can be explained by stock holdings characteristics. Using
different measures for stock valuation, we nd a positive
relationship between the stock-level disposition effect
and valuation uncertainty. Finding that the disposition
effect is more present among certain stocks, we also
investigate the relationship between the analyzed bias
and risk. We nd that the level of the disposition effect is
monotonically increasing with the level of systematic risk
(i.e., beta). Furthermore, we document that the dividend
yield and a dividend dummy have a pronounced positive
effect on mutual fund managers bias, conrming that
not only individual investors but also professional investors show trading behavior motivated by attentiongrabbing stocks. Moreover, our estimates imply that a
reduction in the book-to-market ratio results in a
decrease in the magnitude of the disposition bias. Thus,
fund managers with a higher disposition level tend to
invest in more expensive equities as measured by the
book per price value. The ratio of intangible assets to

total assets as well as earnings volatility have a signicant
negative inuence on the level of the disposition effect.
The results show that fund managers display a higher
level of the disposition effect when the respective stock is
more difcult to value.
Our focus is on fund managers holdings and is motivated by several reasons. First, mutual funds offer a particularly suitable setting for studying the drivers and
implications of organizational decision-making processes. Since mutual funds are subject of regulated disclosure, reliable data are available. Portfolio holdings of
mutual funds offer a rich information set and is suitable
to study information processing and decision making of
mutual fund managers. Second, as professional investors,
mutual funds constantly trade securities in nancial markets and thereby, continuously acquire experience.
Hence, mutual fund managers should on average be
more skilled than retail investors and be more likely to
avoid behavioral biases (Seru, Shumway, and Stoffman
[2010], Cici [2012]). Finally, the mutual fund industry is
of great economic relevance. Given the predominance of
institutional investors in the markets, evidence of the
existence of the disposition effect would have far reaching consequences in conrming or denying the extent of
rationality reected in asset pricing models and related
phenomenon such as the momentum effect. To the best
of our knowledge, our paper is the rst to focus exclusively on identifying stock-related drivers to explain the
cross-sectional variation in the level of the disposition
effect among U.S. mutual fund managers. However, a
number of papers in the broader literature on mutual
fund manager behavioral biases and their holdings contain related results. Busse and Tong [2012] nd evidence
that mutual fund managers tend to pick stocks rather
based on their industry than on rm-specic measures.
Our study uses the variation of the disposition bias
within funds at a given point in time to analyze different
stock-related characteristics. We also document the differences in the level of the disposition effect between
easy-to-value and difcult-to-value stocks.
Our paper relates to the growing body of research
seeking to analyze the role of behavioral biases among
mutual fund managers. Various behavioral biases, such
as overcondence (e.g., Eshraghi and Tafer [2012],
Puetz and Ruenzi [2011], OConnell and Teo [2009]),
anchoring effects (e.g., Kaustia, Alho, and Puttonen
[2008]), and loss aversion (e.g., Alevy, Haigh, and List
[2007], Coval and Shumway [2005]), are explored in previous papers. In the context of behavioral biases among
mutual fund managers, the disposition effect is also
documented in their stock trading activity (Wermers
[2003], Scherbina and Jin [2011], Frazzini [2006], Cici
[2010]). For example, Wermerss [2003] ndings suggest

a reluctance of selling loser stocks among mutual fund
managers in the United States from 1975 to 1994. His
study on the persistence of mutual fund returns conrms
the disposition-prone behavior of mutual fund managers
for winning and losing portfolios. Frazzini [2006] documents the disposition effect after earnings announcements and nds that the extent of the disposition effect
adversely affects returns. Scherbina and Jin [2011]
recently discover the existence of the disposition effect
following instances of managerial change. While existing
managers continue to hold on to their losing investments, newly hired managers do not display the disposition effect as they do not feel responsible for the
inherited mistakes of their predecessors. Ringov [2012]
nds that the extent of the disposition effect is inuenced
by the organizational decision-making process. In contrast, Cici [2010] nds that the average mutual fund
manager is not disposition-prone, and Singal and Xu
[2011] show that only 30% of the mutual funds in their
sample display a disposition tendency. We contribute to
the empirical literature on the presence of behavioral
biases among mutual fund managers. Since mutual fund
managers control a large portion of the U.S. stock market, the presence of the disposition effect may have signicant implications for asset pricing. Perhaps
unsurprisingly, given the perceived opinion that professional investors are the smart money (Frazzini and
Lamont [2008]), empirical research on biases among
professionals has so far reported mixed results. Goetzmann and Massa [2008], for example, report that the disposition effect impacts stock prices; Frazzini [2006]
shows that the disposition effects is related to postannouncement price drift and that it might drive stock
price underreaction to news. A recent paper by Cici
[2012] summarizes the current state of research: I
observe a great deal of heterogeneity among my sample
funds, as 2255% of them exhibit a propensity to realize
gains more readily than losses. While this pattern is consistent with the disposition effect for this subset of funds,
it could also be caused by random variation in the empirical distribution (Cici [2012], p. 796). This statement
combined with the mixed evidence of previous papers
calls for a more in-depth analysis of the disposition effect
among mutual fund managers and its underlying drivers.
We contribute to this literature by providing a
detailed analysis of the disposition effect among a large
and comprehensive sample of actively managed U.S.
funds over a time period of 30 years. By employing the
standard approach of calculating the disposition effect
(Odean [1998]), our results are directly comparable to
previous results. We also add to the research stream analyzing what factors amplify the presence of behavioral
101
biases. In order to understand which factors support or

depress the level of the disposition effect, we study stock
characteristics as one major driving force. To our best
knowledge, there is only one paper, by Kumar [2009],
studying the inuence of stock holdings characteristics
on the level of the disposition bias for a sample of individual investors in the United States from 1991 to 1996.
Recently, Busse and Tong [2012] nd that mutual fund
managers tend to pick stocks based on their industry and
that industry-selection skill drives persistence in performance. We contribute to the literature by analyzing the
trading decisions and related stock holding positions of
professional investors. We provide new insights in the
drivers of the disposition effect by investigating the link
between the disposition effect and various stock characteristics such as rm size, rm age, industry classication, market capitalization, or price-earnings (P/E) ratio.
The rest of the paper is structured as follows. The second section describes related literature. The third sections presents the data. In the fourth section, we provide
details on the employed methodology. We present the
main empirical results in the fth section. In the sixth
section, we give a brief summary and conclude.
Related literature
Since Shefrin and Statman [1985] showed that investors
tend to sell winner stocks too early and hold on to loser
stocks for too long (which they termed the disposition
effect), the effect was conrmed for a variety of investor
types, countries, and asset classes. For example, Odean
[1998], Grinblatt and Keloharju [2001], and Shapira and
Venezia [2001] show evidence of the disposition effect
affecting both inexperienced and sophisticated individual
investors in the United States (Locke and Mann [1999],
Coval and Shumway [2005], Odean [1998]), Finland
(Grinblatt and Keloharju [2001]), and Israel (Shapira
and Venezia [2001]), respectively. Evidence of disposition effects is also uncovered in other asset classes, such
as residential housing (Genesove and Mayer [2001]),
executive stock options (Heath, Huddart, and Lang
[1999]), and prediction markets (Hartzmark and Solomon [2012]). Among different explanations suggested
for the observed behavior, the disposition effect is attributed to key features of mental accounting (Thaler
[1985]) and prospect theory (Kahneman and Tversky
[1979]). People tend to value gains and losses relative to
a reference point (e.g., purchase price in Odean [1998]).
Prospect theory refers to a risk averse preference in the
domain of gains and risk seeking in the domain of losses
(both measured relative to a reference point). The main
idea of mental accounting is that investors set different
reference points for different accounts which determine
102
S. WULFMEYER
gains and losses. They divide different types of gambles

into separate accounts, and use prospect theory for each
account separately, thereby ignoring possible interactions. The combination of mental accounting and prospect theory creates the disposition effect.
At an institutional level, the existence of behavioral
biases is less clear. Given their predominance in the market, evidence of irrational behavior would have far reaching consequences in establishing or denying the market
efciency hypothesis. Locke and Mann (1999) nd evidence of the disposition effect among future trader using
transaction data from the Chicago Mercantile Exchange
(CME). Coval and Shumway (2005) show that the disposition effect exists among a group of propriety traders at
the Chicago Board of Trade. Shapira and Venezia (2001)
prove the disposition effect among professional investors
in Israel. The disposition effect is also found among U.S.
equity mutual fund managers (e.g., Wermers [2003],
Frazzini [2006], and Scherbina and Jin [2011]). Hence,
better decision making and the ability to create superior
performance of institutional investors remain questionable. Wermers [2003] documents that loser funds tend
to be more disposition-prone and that managers of
underperforming funds appear reluctant to close their
losing positions. Contrary, successful managers realize
losses at a higher rate than gains. Frazzini (2006) conrms previous results documenting the disposition effect
after earnings announcements. In their study about
newly hired fund managers, Scherbina and Jin [2011]
conclude that previous managers are reluctant to sell
loser stocks in their portfolios. Other recent studies conrm previous results and show that U.S. equity mutual
fund managers are disposition-prone (e.g., Cici [2010],
Ammann, Ising, and Kessler [2012], Ringov [2012], and
Chiang and Huang [2010]).
Although the disposition effect is documented for a
variety of settings, time periods and assets, its underlying
drivers remain rather unclear. Some papers argue that
fund characteristics (e.g., fund costs, turnover, fund volatility, fund size, funds age) determine the level of the disposition effect (Cici [2012], Solomon, Soltes, and Sosyura
[2012]). Other papers nd fund managers characteristics
(e.g., age, experience, team- or single managed fund) to
infer with the level of disposition-prone behavior (Scherbina and Jin [2011], Solomon, Soltes, and Sosyura
[2012], Ringov [2012]). Recently, the disposition effect
has been attributed to fund in- and outows (Cici
[2010], Chiang and Huang [2010], Singal and Xu
[2011]). Stock characteristics are identied as another
possible driver of the disposition effect. For example,
Frazzini [2006] nds that purchases and sales of attention-grabbing stocks inuences the disposition effect.
Cici [2010] documents the disposition effect among
loser funds and investors who systematically add loser

stocks to their portfolio in what appears to be some
kind of gambling with losses. So far, there is only one
paper (by Kumar [2009]) that analyzes stock characteristics such as rm size, rm age, market capitalization, P/E
ratio as drivers of the disposition effect for retail investors. Recently, a paper by Busse and Tong [2012] shows
that mutual fund managers tend to pick stocks based on
their industry. We add to the research stream of stockrelated drivers of the behavioral biases by analyzing how
various stock holdings characteristics inuence the level
of the disposition effect.
Data
Data sources and sample construction
The data used in this study come from four main sources: quarterly fund holdings data from Thomson Financial (previously known as CDA/Spectrum), monthly and
annual mutual fund characteristics from Center for
Research in Security Prices at the University of Chicago
(CRSP), daily and monthly stock return les from CRSP,
and monthly and annual accounting data from Compustat. We obtain our main fund sample by merging the
CRSP Survivorship Bias Free Mutual Fund Database
(CRSP MF, henceforth) and the Thomson Reuters
Mutual Fund Holdings database (TR, henceforth) using
MFLinks from Wharton Research Data Services
(WRDS). For each single fund, information about the
fund characteristics (e.g., sector, style, starting date, manager), performance information (e.g., returns, asset
under management, fees) is extracted from CRSP.
In addition to the fund characteristics from CRSP, we
extract holdings information from the TR database. The
TR database reports all changes in holdings as well as
holdings characteristics, that is, ticker symbol, permno
(CRSPs permanent stock issue identier), cusip (CRSPs
stock identier), and the price of each asset on a quarterly basis. The cusips are used to extract different
accounting information and trading statistics for each
stock from Compustat. The price for each stock is taken
at the date of reporting as well as at the end of each
month as in previous literature (e.g., Frazzini [2006],
Barber and Odean [2008]). Furthermore, we get daily
data from CRSP to calculate stock-related variables, for
example, stock volatilities, betas, market-to-book ratio,
and returns. Data on the annual book value of common
equity, cash ows, market capitalization, intangible
assets, earnings, dividend payments, PPE (property,
plant, and equipment), and total assets are retrieved
from Compustat. Information on analyst coverage is
downloaded from I/B/E/S.
To arrive at the nal sample used in the empirical

analysis, we start with the entire universe of U.S. mutual
funds for the period from 1980 to 2010. Next, following
previous literature, we limit our sample to actively managed, diversied equity mutual funds which are based in
the United States3. We match the TR and CRSP mutual
fund datasets for the period from 1980 to 2010.
In our sample, there are funds with multiple asset
classes holding the same portfolio of stocks since they
are listed as separate entities in the CRSP database. They
usually vary with respect to their fee structure or minimum purchase limits but are based on exactly the same
portfolio of assets. To avoid multiple counting, we aggregate all share classes of the same fund using the unique
WFICN (Wharton Financial Institution Center) fund
number to aggregate fund data across different share
classes into one observation per fund-year. For variables
which vary across share classes (e.g., returns, turnover
ratio, expense ratio, etc.), we take weighted averages
using total net assets as weights.
For benchmarking our results, we use the return on
the S&P 500 index as the market portfolio and the 3month Treasury bill rate as the risk-free rate, which are
provided by Datastream.
Control variables
In order to avoid spurious correlations, we control for
the effect of variables that might inuence the level of a
funds disposition effect while being correlated with the
independent variables of interest in this paper. We control for different fund-related characteristics such as
fund size, fund age, fund costs, a funds capacity, turnover, expense ratio, institutional or retail ownership, and
fund performance.
We also control for fund manager characteristics since
they may also inuence the level of the disposition effect.
We control for a fund managers experience, a new manager dummy variable, and whether a fund is team- or
single-managed. In our regression settings, we also
include several stock-related control variables. We
include rm age and rm size to control for various sizerelated effects in our stock-related regression specications. Turnover is included to account for possible
liquidity-based effects. We control for monthly past
returns since the overall tendency to realize capital losses
more readily than capital gains could also be due to following a momentum strategy by mutual fund managers.
The past return variable serves as a control variable for
the known effects of past returns on the trading behavior
of investors (for retail investors see Barber and Odean
[2008]). For instance, the trading activities of mutual
fund managers might be inuenced by following a
103
contrarian strategy or trend-following behavior. Past

returns may additionally capture other kinds of behavior
such as rebalancing. Furthermore, Odean [1998] mentions that past returns could also reect transaction costs
consideration. For example, funds which have to sell due
to liquidity reasons might prefer to sell a stock that has
recently increased in value since stocks which decreased
are lower priced and hence are less liquid (Odean
[1998]).
We also control for stable unobservable effects inuencing the level of the disposition effect by including
xed effects for each fund-year combination. All control
variables except for the xed effects are lagged one
period (year). For a detailed description and denition of
the variables please refer to Appendix B.
Table 1. Summary Statistics of Fund Characteristics. This table

reports the summary statistics of fund characteristics of all mutual
funds in the sample over the period from January 1980 through
December 2010. The table displays the number of observations
(N), mean, standard deviation, minimum and maximum of the
different fund-related variables. Fund age is measured as the
years between the rst offer date and the last reporting date.
Fund size is measured as quarterly total net asset value in millions of dollars. Monthly TNA is the monthly total net asset value
calculated in millions. Monthly NAV is the monthly net asset
value measured in millions. Managers experience is measured in
years that the manager has worked at a specic fund. The
dummy variable new manager is equal to one if the manager is
less than three years at the specic fund or zero otherwise. Fund
capacity is a dummy variable which is equal to one if the fund is
open to investments and zero otherwise. The variable institutional dummy is equal to one if the fund is an institutional fund
and zero otherwise. The retail dummy variable is equal to one if
the fund is a retail fund and zero otherwise. Actual 12b-1 fees are
reported as the ratio of total assets attributed to marketing and
distribution costs. Maximum 12b-1 fees report the maximum of
the actual 12b-1 fees. Front load for investments represents maximum sales charge at breakpoint. Rear Load are fees which are
charged when withdrawing funds.
Variable
Fund age
Fund size
Monthly TNA
Monthly return
Monthly NAV
Manager experience
New manager
Fund capacity
Institutional dummy
Retail dummy
Actual 12b-1 fee
Maximum 12b-1 fee
Front load
Rear load
Expense ratio
Turnover ratio
Return
Funds
Fundyear
Mean
Std. Dev
2260 535 12.54 9.9363

3.95
2.2947
3820 097
1960 673 572.22 20 739.52
0
198 331 0.0082 0.0633
1980 364 15.3728 14.3328
1460 649 10.26 5.57
0.03
0.17
3820 097
2930 408 0.8909 0.3117
2930 408 0.3300 0.4702
2930 408 0.6637 0.4724
1450 402 0.0054 0.0036
1410 363 0.0058 0.0034
1910 074 0.0058 0.0145
2200 945 0.0027 0.0087
2700 523 0.0134 0.0073
2690 078 0.8982 0.8608
3810 301 0.0011 0.1267
20 605
750 545
Min
Max
0
6.9078
0.10
0.3112
0.1600
1.99
0
0
0
0
0
0.0005
0
0
0.0051
0
0.8776
86
11.5998
580 394.60
0.5097
411.6600
54.03
1
1
1
1
0.0345
0.0100
0.1360
0.0600
0.4845
29.5200
4.0476
104
S. WULFMEYER
Sample characteristics
Table 1 presents the summary statistics for the 2,605
funds in our sample for the time period 19802010. The
sample consists of 2,605 funds and 75,545 fund-year
observations. Our sample starts in 1980 with 8 funds and
consists of 1,185 funds in 2010. Monthly total net assets
increase from an average of $227 million in 1980 to
$720 million in 2010. The average fund in our sample is
around 13 years in business with an average fund size of
$ 4 billion. The turnover ratio at the beginning of the
sample period is 77%, which is equal to a holding period
of 1.5 years. While there is signicant variation in the
expense ratio over time, the mean turnover for the whole
sample period is 86%, implying a holding period of
1.2 years. The turnover gures are comparable to those
reported in prior research of between 85% and 100%
(e.g., Singal and Xu [2011], Cici [2010], Kacperczyk,
Sialm, and Zheng [2008]). The expense ratio was 1.26%
in 1980 and increased over time to 1.3% in 2010. The
highest expense ratio was 1.8% in 1985, and the lowest
expense ratio was documented at 0.7% in 1996. Over the
last 30 years, the average mutual fund had a return of
0.82%, with the highest return of 13.1% in 1991. We
document that the average fund charges a front load of
0.6%, paid upfront by the fund investors. The average
rear load is 0.3%, which is charged when withdrawing
the money from the respective fund.
Most of the funds are managed by teams with an average experience of a fund manager of 10 years. We also
report whether a fund is taken over by a new fund manager (dened as three years or less at a specic fund); we
nd that only 3% of the funds are managed by new fund
managers. We also look at a funds capacity which is a
dummy variable equal to one if the fund is open to new
investors and zero otherwise. We nd that mutual funds
are on average open to investors (89%) and that only
around 10% of the funds are closed.
We also look at the summary statistics for the number
of shares and the value of trades in our sample. Table 2
shows the trade statistics for the 2,605 mutual funds in
our nal sample. The fund managers purchased about
180,000 stocks and sold 195,000 stocks during the
Table 2. Statistics of Trading Activity
Variable
Buy
Sell
Total
sample period from rst-quarter 1980 to last-quarter

2010. Overall, our sample consists of 373,610 changes in
assets. On average, there were 6,000 purchases and 6,500
sales per year, which means that each fund has roughly 5
trades per year that are reported to the SEC on a quarterly basis. Table 3 displays the value of the trades of all
mutual funds during the sample period from 1980 to
2010. The average market value of all purchase is
$3 million and $10 million for sales.
Methodology
We calculate the disposition effect following the method
of Odean [1998] in order to determine how quickly fund
managers realize capital gains as well as capital losses.
We measure the extent to which mutual fund managers
behavior exhibits the disposition effect by the disposition
spread. To determine the disposition effect we use stock
holdings information of mutual funds as well as the
weighted average purchase price. For any given investor,
the proportion of all potential realized gains (PGR) and
realized losses (PLR) is calculated and compared to paper
gains and losses, respectively.
Our research design is implemented as follows: each
quarter that a sale takes place, the selling price is compared to a reference price in order to determine whether
the sale is a realized gain or loss. In order to determine
the reference price, one has to be clear about which cost
basis to use. In our paper, the reference price is the historical weighted average purchase price (WAPP), which
is updated each time a buy transaction takes place4.
Using the WAPP as reference price is based on the
assumption that fund managers regularly update their
reference points after each (net) purchase. The WAPP
for stock i hold by fund x on a day t is calculated as
follows
WAPPi;x;t D SharePrci;t SharesBuyi;t
C WAPPi;x;t 1 SharesHeld i;x;t 1
6 .Shares Held/i;x;t
(1)
Table 3. Statistics for Value of Trades
Mean
Std. Dev
Min
Max
1780 987
1940 623
3730 610
1080 095
1010 170
917
5290 138
5470 238
5480 693
1
53.5E C 07
53.5E C 07
26.9EC07
1
26.9EC07
This table reports the summary statistics for the number of shares during the
sample period from 1980 to 2010 traded per quarter. The table shows the
number of purchases and the number of sales as well as total trades. Purchases are dened as changes in holdings which are larger than zero. A sale
is dened as a change in stock holdings being equal or smaller than zero.
Variable
Buy
Sell
Total
Mean
Std. Dev
Min
Max
1780 987
1940 623
3730 610
20 9690 60
1010 170
1580 336
0 1.3EC07
5470 238
1.96EC07
1
5.3EC07
5.6EC09
5.76EC08
1
5.76EC8
This table reports the summary statistics for the value of purchases and sales
in dollars traded during the sample period from 1980 to 2010 in the
reported holdings of U.S. equity mutual funds. Purchases are dened as
changes in holdings which are larger than zero. A sale is dened as a
change in stock holdings being equal or smaller than zero.
where SharePrc is the respective share price, SharesBuy

indicates the number of shares bought, and SharesHeld is
dened by the number of shares held in the portfolio. We
have to distinguish two different cases: whether the fund
decreases or maintains the level of its holdings of stock i,
then the WAPP of stock i on day t equals the WAPP of
stock i on day (t 1). In the second case, the fund
increases its level of holdings of stock i, then the WAPP
of stock i on day t held by fund x is calculated as in
Equation 1. The calculation of the WAPP based on the
above equation is done for each fund-stock-date
combination.
Due to limitations of the data and to be consistent
with other studies (e.g., Cici [2010], Huang, Wei, and
Yan [2007], Frazzini [2006]), we assume that all changes
in holding occur at the end of each quarter5. By a comparison of holdings at the end of one quarter with holdings at the end of the previous quarter, we are able to
determine whether a sell or buy transaction has taken
place during the respective quarter. On each day a sale
takes place, the selling price for each stock is compared
with the WAPP in order to determine whether the sale
transaction is a realized gain (RG) or loss (RL). For all
other stocks in the portfolio of a specic fund which are
held but not sold on the same day as a given stock is
sold, the market price is compared with the WAPP in
order to determine whether it is a paper gain (PG) or
paper loss (PL).
Having calculated the RG, RL, PG, and PL, we are
now able to calculate the proportion of realized gains
(PGR) and realized losses (PLR) on each day that a fund
is a net seller of a certain stock. The following equations
show the calculation of the PGR and PLR:
PGRi;t D
realized gainsi;t
realized gainsi;t C paper gainsi;t
realized lossesi;t
PLRi;t D
realized gainsi;t C paper gainsi;t
(2)
(3)
We further calculate the disposition spread as difference between proportion of gains realized and the proportion of losses realized by a mutual fund in a given
period (e.g., Dhar and Zhu [2006], Frazzini [2006],
Goetzmann and Massa [2008], Kumar [2009], Odean
[1998], Ringov [2012]).
DSi;t D PGRi;t PLRi;t
(4)
If the disposition spread is positive, then the respective

fund realizes disproportionally more gains than losses
(disposition-prone fund). Therefore, the larger the disposition spread gets, the stronger is the level of the
105
disposition effect exhibited by the respective fund manager or team of fund managers. We measure the average
quarterly disposition effect for each fund by aggregating
the number of gains and losses by each fund and taking
the average (see Chiang and Huang [2010]). The annual
disposition effect (which is used in the regression settings)
is calculated as the average of the four quarters of each
year.
Empirical analysis
Disposition effect
Level of the disposition effect
In this section, we look at the disposition spread calculated as the proportion of realized gains (PGR) minus
the proportion of realized losses (PLR). The disposition
spread of all funds in our sample for the entire sample
period from 1980 to 2010 is shown in Panel A of Table 4.
Results are based on the assumption that trades happen
sometime during the quarter and hence, averages of daily
stock prices during the respective quarter are used. The
PGR, PLR and disposition spread measures are rst
Table 4. Disposition EffectSummary Statistics
Panel A
Period
19801989
19901999
20002010
Total
N
810 186
5090 541
10 2560 718
10 8470 445
Period
N
Non-crisis 10 0500 130
Crisis
7970 315
Total
10 8470 445
% of funds Mean Std. Dev p25

4.39
27.58
68.02
100
0.0464
0.0448
0.0370
0.0396
Panel B
% of funds Mean
56.84
0.0396
43.16
0.0395
100
0.0396
p50
p75
0.0053
0.0064
0.0084
0.0086
0.0441
0.0414
0.0324
0.0348
0.0454
0.0444
0.0371
0.0399
0.0474
0.0469
0.0422
0.0449
Std. Dev
0.0084
0.0088
0.0086
p25
0.0353
0.0342
0.0348
p50
0.0400
0.0397
0.0399
p75
0.0450
0.0448
0.0449
Panel A of this table reports the disposition effect. It reports the number of
observations (N), the percentage of funds, mean, median, the 25th and 75th
percentiles as well as the standard deviation (Sd) of the disposition effect
for the whole period and for different subperiods 19801989, 19901999
and 20002010. The disposition effect is calculated as the proportion of
gains realized minus the proportion of losses realized. The proportion of
gains realized (PGR) is dened as the ratio of realized gains to the sum of
realized and unrealized gains. Accordingly, the proportion of losses realized
(PLR) is measured as the ratio of realized losses to the sum of realized and
unrealized losses. The table displays the absolute share numbers for all
funds in our sample. The underlying assumption for the calculation of (capital) gains and losses is that mutual fund managers rst sell those stocks with
the highest cost basis (HIFO method). PGR, PLR and DS are calculated on a
quarterly and on fund level basis. The numbers are calculated in a two-step
process: rst, the mean PGR, PLR, and DS for each fund is calculated for all
quarters in which the fund has valid data, and second, the mean PGR, PLR
and DS across all funds are averaged. We show results for all four quarters
measured in number of trades. Panel B reports the disposition results for
crisis and noncrisis periods. We dene the following crisis times: the US
recessions from 1980 to 1982 and from 1990 to 1992, the Asian crisis of
1997, the Russian crisis 1998 and the collapse of Long-Term Capital Management (LTCM) in 1998, the Dot-com bubble from 2000 to 2001 and the
recent nancial crisis from 2007 to 2009. We report the level of the disposition effect, the number of observations (N) and the percentage of funds.
106
S. WULFMEYER
calculated for each single fund separately. Second, we

take the time-series average of the fund-level measures to
calculate the overall disposition spread. In doing so, we
assume independence of the disposition effect across
mutual funds.
For the results of the whole sample, there is a signicantly positive disposition spread (3.96%) between
PGR and PLR indicating that on average mutual
funds in our sample are prone to the disposition bias.
A large fraction of funds shows a positive disposition
spread, which means that the majority of funds in
our sample display the analyzed behavioral bias.
Table 4 shows a decline of the level of the disposition
effect over time. While in 1980 the bias has a value
of around 4.6%, it is around 4.4% in 1990 and falls to
3.4% in 2010. This pattern might be due to the
increased awareness of behavioral nance among
mutual fund managers. Some of the rst research on
the disposition effect was published during our second subperiod (e.g., Odean [1998], Weber and
Camerer [1998], Grinblatt and Keloharju [2001]).
Thus, consistent with practitioners rst being introduced to behavioral biases in the 1990s, we observe
that the level of the disposition effect steadily declines
thereafter.
Our results are in line with previous research. A
paper by Rosa, To, and Walter [2005] nds an average disposition effect of 0.12% ($-value basis: 0.0559)
for a sample of U.K.-managed funds. Other papers
also nd a positive disposition spread (e.g., Frazzini
[2006], Odean [1998], Barber and Odean [2008]) and
further conrm our results. Frazzini [2006] nds a
positive disposition spread (3.1%) for professional
investors in the United States. In comparison, Odean
[1998] documented a disposition effect of 0.016 using
share-based numbers for individual investors. In line
with this, a recent paper by Ammann, Ising, and
Kessler [2012] shows that the disposition effect is
3.2% calculated on a share basis for 19932005. Our
results are also conform to Scherbina and Jins [2011]
paper, who report that old managers are reluctant to
realize capital losses compared to their newly hired
counterparts, thereby supporting the disposition
effect.
New explanations and models based on private information are documented in the recent papers by BenDavid and Hirshleifer [2012] and Dorn and Strobl
[2011]. A leading explanation for the presence of the disposition bias is that investors are reluctant to realize
losses because of a direct disutility from doing so. For a
sample of individual U.S. investors, Ben-David and
Hirshleifer [2012] document the opposite, nding that
the disposition effect is not driven by a simple direct
preference for selling a stock by virtue of having a gain

versus a loss but rather by beliefs. In line, Dorn and
Strobls [2011] theoretical model, which is based on the
information asymmetry between informed and uninformed investors, attributes the disposition effect to
investor preferences rather than beliefs.
In sum, we document that the average mutual fund
has a stronger preference to lock in gains than to realize
losses. Although the average spread level is positive, we
observe some cross-sectional variation in the range of
the effect. Table 4 shows that the disposition spread has
a standard deviation of 6.3%. Therefore, this paper is
devoted to shed new light on evaluating which stocks
mutual fund managers trade and whether there is systematic variation between disposition-prone and lessdisposition-prone mutual funds.
Results stratied by subperiod
In this section, we focus on the question whether the
behavior of mutual funds with respect to the realization
of gains and losses has changed over time. We are especially interested whether there are certain learning effects
when looking at the level of the disposition effect over
time. Table 4 reports the results for the whole period and
for subperiods, namely 19801989, 19901999, and
20002010. On average, mutual funds show the disposition effect in any of the subperiods. In 19801989 and
from 19901999, funds show an increased tendency to
realize disproportionally more gains than losses.
Especially since the beginning of the 1990s, the level of
the disposition effect sharply declines. This pattern
might be due to the increased awareness of behavioral
nance since the 1990s among mutual fund managers,
so the level of the disposition effect steadily declines
since then. In addition, advances in technology affecting
quantitative investment strategies could have introduced
more discipline and less reliance on biased human decision making.
A second explanation for the observed behavior is that
possible learning effects of mutual fund managers affect
our results. A reduction in the level of the disposition
effect through time is consistent with the theory that
investors learn from trading. Informed investors update
their price targets more often and hence, are less prone
to display behavioral biases. Kumar and Lim [2008] document that investors who tend to execute several trades
during the same day suffer less from the disposition
effect. Previous ndings suggest that more active fund
managers have persistent stock-picking skills and learn
to avoid biases over time (Wermers [2000]). This might
be a rst indication that mutual fund managers update
their price targets regularly and learn to avoid behavioral
biases through time.
Disposition effect and stock characteristics

Baseline results
In the previous section, we show the existence of the
disposition effect among U.S. mutual fund managers
from 1980 to 2010 and how fund characteristics inuence the magnitude of the observed behavioral bias.
The documentation of the disposition effect raises the
questions of the driving forces behind the effect. In
the previous section, we nd various fund characteristics as being inuential on the likelihood of realizing
a larger proportion of winner stocks than their losing
counterparts. We also established that there is dispersion among the level of the disposition effect, which
further raises the question whether the disposition
effect is linked to differences in investment styles. In
this section, we study the impact of different stock
holdings characteristics on the likelihood of a fund
to show a positive disposition effect.
To analyze whether mutual fund managers with different levels of the disposition effect follow different
investment approaches, we investigate the characteristics
of their stock holdings. We analyze different stock characteristics like betas, trade volume, Sharpe ratio, market
capitalization, mean return, volatility, etc. of the stocks
held by mutual funds. We obtain the stock characteristics
for each mutual fund by taking value-weighted averages
of the single stock gures (for further details on how
each variable is constructed, please refer to Appendix B
Table B.3).
In Table 5 we report the summary statistics for the
stock holdings of the 2,605 mutual funds in our sample
from 1980 to 2010. The 2,605 funds hold 15,110 different
stock positions (identied by their individual cusips during the sample period from 1980 to 2010). To measure
company-specic differences, we analyze the mean ratio
of intangible assets to total assets which is roughly 10%.
Since the mean book-to-market ratio (0.52) is varying
signicantly (standard deviation of almost 100%), we
use the median book- to-market value (0.44) in our
regression settings. Earnings per share (EPS) are on
average 0.73, whereas the P/E ratio has a mean of 13.11.
We also include EPS as another nancial ratio, which is
the dollar value of earnings per outstanding share of a
rm's common stocks. As stock characteristics, we also
look at different volatility measures, namely earnings,
cash ow and return volatility. In addition, we also
include the dividend yield (mean of 2.32%) and a dividend dummy which is equal to one when the stock is
paying a dividend and zero otherwise. When we analyze
the trading decisions of mutual fund managers, we also
include fund size, market capitalization, and turnover of
the respective stock.
107
Table 5. Summary Statistics of Stock

Variable
Book-to-market ratio
Earnings-per-share
Earnings volatility
Price-earnings ratio
Cash ow volatility
Dividend yield
Dividend dummy
Intangible assets (in%)
Property, plant,
equipment (in %)
Market capitalization
Firm size
Mean stock return
Stock turnover
Shares Outstanding
Volume
Market return
Mean Std. Dev
Min
Max
870 066
870 066
870 066
870 066
870 066
870 066
870 048
870 066
870 066
0.5233
0.7349
176.75
13.11
193.42
0.0232
0.4480
0.1014
0.5207
0.9893
2.6857
784.49
35.95
790.23
0.3279
0.4973
0.1599
0.4183
64.8035
282.8433
0.05
2094.79
0.05
0.0019
0
0.0053
0
6.3293
54.3558
23115.20
1107.91
18386.25
15.5973
1
0.9970
6.8626
1630 628
1630 628
1630 628
1630 628
1630 628
1630 628
1630 628
1.41
12.25
0.2072
6.32
0.05
0.37
0.0874
6.59
1.80
0.2125
22.68
0.22
2.60
0.0474
0
5.48
3.5945
0.01
0
0
0.3282
172.00
18.96
10.6995
1834.38
10.80
300.00
0.3560
CharacteristicsThis table presents the characteristics for the 15,110 different

stock holdings of the mutual funds in our sample from 1980 to 2010.
Accounting and stock-related information are retrieved from Compustat
and CRSP, respectively. The table reports annual estimates based on
monthly data for accounting variables (intangible assets as percentage of
total assets, book-to-market ratio, earnings per share, earnings volatility,
price-earnings ratio, cash ow volatility, dividend yield, property, plant,
equipment as percentage of total assets) and daily time series for all other
variables. For the denition of variables please refer to Appendix B
Table B.3. The table reports the number of observations, mean, standard
deviation, minimum and maximum values. Market capitalization is reported
in millions of dollars; shares outstanding and volume are reported in
millions.
In Table 6 we report the summary statistics for the

stock characteristics for the 10 disposition deciles Deciles
are based on the average disposition spread of each fund
during the whole sample period. Decile one contains the
funds with the lowest disposition spread, whereas decile
ten includes funds with the largest disposition spread. In
this statistics, a fund can only be counted once; if the
database contains more than one report for the respective mutual fund, we calculate the average disposition
spread and the average stock characteristics across time.
The differences in the level of the disposition effect can
be interpreted as an indication of the stock picking ability of a mutual fund manager. Table 6 shows that fund
managers with a higher disposition spread invest in equities with lower average trade volume and lower market
capitalization.
We observe that mutual fund manager with a lower
disposition level invest in different assets than managers
with a higher disposition spread. We show that fund
managers with a lower level of the analyzed behavioral
bias prefer stocks of older and larger rms. In the lowest
disposition decile, the average rm age as measured the
number of years when the stock rst appeared in CRSP
is roughly 18 years compared with 17 years in the highest
disposition decile. The rm size, which is dened as the
natural logarithm of the end-of-quarter market capitalization (shares outstanding times prices), is larger for
108
S. WULFMEYER
Table 6. Sorting Results: Stock Characteristics of Mutual Fund Holdings in Different Disposition Spread Deciles
Disposition
decile
Disposition
spread
Market
capitalization
Firm
size
Stock
turnover
Intangible
assets
Book-tomarket ratio
Earnings-per-share
Priceearnings ratio
Dividend
yield
1
2
3
4
5
6
7
8
9
10
t-statistics
0.0335
0.0362
0.0371
0.0379
0.0385
0.0391
0.0398
0.0406
0.0427
0.0457
4.53
5.76
5.25
5.95
3.68
7.98
13.50
20.20
3.73
1.57
102.74
14.10
14.27
14.33
14.15
14.29
14.62
15.25
15.15
13.66
13.13
196.14
11.37
10.57
9.90
8.50
8.53
7.68
7.02
6.33
6.08
5.96
133.11
0.1977
0.1954
0.2023
0.1494
0.1636
0.1622
0.1482
0.1170
0.1133
0.0979
88.89
0.45
0.42
0.45
0.44
0.43
0.44
0.42
0.41
0.53
0.57
18.65
1.31
1.41
1.38
1.21
1.46
1.59
1.77
1.71
1.17
0.76
38.96
17.94
21.35
17.58
19.15
15.65
20.22
23.93
20.56
13.75
22.01
16.12
0.0249
0.0164
0.0190
0.0105
0.0145
0.0339
0.0333
0.0359
0.0189
0.0139
15.57
The table contains the averages of different characteristics of the mutual fund stock holdings by disposition spread deciles. The disposition effect deciles are calculated over the whole sample period from 1980 to 2010. Decile one contains the funds with the lowest disposition spread, whereas decile 10 included funds
with the largest disposition spread. In this table, a fund can only be counted once; if the database contains more than one report for the respective mutual fund,
we calculate the average disposition spread and the average stock characteristics across time. Market capitalization is reported in millions of dollars. Firm size is
dened as the natural log of market capitalization (in millions), which is calculated as the number of shares outstanding (shrout) times price. Stock turnover is
the ratio of the number of shares traded and shares outstanding. The variable intangible assets is the ratio of intangible assets to total assets. Book-to-market
ratio is calculated as the book value divided by the current market price. Earnings per share is net income divided by common shares outstanding. Price-earnings ratio is the stock price divided by earnings (net income). The reported t-statistics show the signicance of the difference between the rst and tenth deciles
for the respective variable.
lower disposition deciles and decreases monotonically

for the higher disposition deciles. This means that
mutual fund managers with a higher disposition effect
try to achieve a superior performance by investing in
rather smaller stocks. The difference between the highest
and lowest decile for rm age and rm size are highly
signicant (t-statistics of 31.99 and 196.14, respectively).
Kumar [2009] also nds a negative coefcient for the
relationship between rm age and the disposition effect
for a sample of the holdings of individual investors in
the United States from 1991 to 1996. In order to capture
the level of uncertainty in fundamental values of a stock,
we include the following variables: the level of intangible
assets to total assets, cash ow volatility, and earnings
volatility. Concerning the level of intangible assets to
total assets, we nd that less biased fund managers invest
in companies with a high level of intangibles assets
(19.77% for the lowest disposition decile). To account
for stock-level uncertainty, we further calculate cash ow
volatility where cash ows are dened as income before
extraordinary items plus depreciation and amortization.
The variable cash ow volatility does not vary a lot across
the different deciles and it seems that there are no signicant differences between the highest and lowest disposition decile. The empirical results from the uncertainty
measure (cash ow volatility) indicates that mutual fund
managers exhibit a stronger level of the disposition effect
and hence, make larger investment mistakes, when
stocks are more difcult to value. Contrary, when the
level of intangible assets is high (which means higher
uncertainty), the analyzed behavioral bias is less prone.
The results concerning cash ow are in line with previous ndings by Kumar [2009] documenting that if fundamentals-based uncertainty measures are high, the
behavioral biases are stronger. Hence, his empirical

results indicate that individual investors (large U.S. sample of individual investors from 1991 to 1996) make
larger investment mistakes when stocks are more difcult to value.
When analyzing the book-to-market ratio, we nd
that it is lower for disposition decile 1 (value of 0.4454)
compared with decile 10 (value of 0.5656), which means
that less biased fund managers tend to invest in more
expensive equities as measure by the price to book value.
Previous results by Kumar [2009] conrm that behavioral biases (in his case: overcondence) is more present
among stocks which have lower book-to-market ratios.
Our sorting results for the variable EPS indicate that less
biased fund managers trade stocks with higher EPS. We
nd that fund managers in the highest disposition decile
invest in stocks with a signicantly lower EPS ratio compared with their less biased counterparts. More biased
fund managers seem to choose on average less protable
companies as measured by EPS. Since funds displaying
the disposition bias perform worse than their unbiased
counterparts (e.g., Singal and Xu [2011], Cici [2012]),
our nding that biased fund managers tend to invest in
less protable stocks may provide an explanation.
We also analyze the relationship between the disposition effect and the P/E ratio. For fund managers in the
highest decile, we measure a higher P/E ratio of 22.01
than for the lowest deciles with a P/E ratio 17.94 (t-statistics of 16.12). Therefore, less biased fund managers
seem to invest in cheaper stocks as measured by the P/E
ratio. The empirical evidence is not consistent with
results of Odean [1998] for a sample of U.S. individual
investors from 1991 to 1996, who nds that the disposition effect is stronger among lower-priced stocks. In
general, a higher P/E ratio suggests that the market

expects a higher earnings growth in the future compared
to companies with a lower P/E ratio. Hence, more biased
fund managers seem to be more optimistic about the
future by investing in high P/E stocks.
In summary, we nd that there is a relationship
between the disposition effect displayed by mutual fund
managers and the characteristics of their portfolio holdings. Our results indicate that mutual fund managers
with a lower disposition effect level invest in different
assets than their more biased counterparts. Less disposition-prone investors pick stocks of companies with a
smaller market capitalization, with more investments in
intangible assets, and a lower book-to-market ratio. Portfolio holdings of less disposition-prone fund managers
also tend to have a lower EPS and P/E ratio. We also nd
that stocks with a higher turnover and a higher dividend
yield are preferred by less biased fund managers. In sum,
we document that stock characteristics of mutual fund
holdings explain part of the cross-sectional variation in
the level of the disposition effect.
Regression-based evidence
To gain more insights into the relationship between the
level of the disposition effect and stock characteristics,
we estimate different regressions. This section further
analyzes the link between the disposition effect and stock
characteristics by employing a regression setup. We
employ the following regression specication:
DS
DS D a C bDS
k Stock Characteristicsk C e
(5)
where stock characteristics are book-to-market ratio, P/E

ratio, EPS, property plant and equipment (PPE) in percentage of total assets, intangibles assets as percentage of
total assets, earnings volatility, cash ow volatility, dividend yield, and a dividend dummy being equal to one if
the stock is paying a dividend and zero otherwise (for
the denition of these variables please refer to
Appendix B Tables B.3).
In Tables 7 and 8 we present the results of the regressions using different OLS regression models with yearfund xed effects. Model 1 to 4 of Table 7 include several
control variables as well as different xed effects. To
account for potential serial and cross-sectional correlations in the error terms, we compute rm-clustered and
year-clustered standard errors. In all models, the results
suggest a highly signicant positive relationship between
the level of the disposition effect and stock characteristics. Overall, the results show that part of the differences
in the disposition effect can be explained by differences
in stock holdings (R2 between 11.06% and 12.98
109
Table 7. Disposition Effect and Accounting-Related Stock Characteristics: Regression Estimates

Model 1
Stock-level DE
Model 2
Model 3
0.0401
0.0397
0.0392
(90.09)
(88.86)
(88.03)
Book-to-market ratio 0.0011
(29.73)
Price-earnings ratio 0.0000
(16.72)
Earnings-per-share
0.0000
(7.77)
PPE
0.0001
(5.23)
Intangible assets
0.0003
(6.55)
Earnings volatility
0.0000
(11.75)
Cash ow volatility
0.0000
(11.33)
Dividend yield
0.0004
(8.60)
Dividend dummy
0.0005
(24.98)
950 278
950 277
N. of cases
950 278
R-squared
0.1201
0.1106
0.1164
Model 4
0.0398
(89.91)
0.0012
(31.70)
0.0000
(13.52)
0.0000
(0.53)
0.0000
(0.20)
0.0003
(6.19)
0.0000
(12.42)
0.0000
(4.95)
0.0007
(14.43)
0.0006
(27.27)
950 277
0.1298
This table reports the regression estimates where the disposition spread in a
given stock in a year is employed as dependent variable. The disposition
effect is rst calculated on a quarterly basis for each fund, secondly, aggregated to annual measures and thirdly, summarized for each stock in the
holdings of mutual funds during our sample period from 1980 to 2010. We
include the following accounting-related stock variables in our regressions:
book-to-market ratio, price-earnings ratio, earnings per share, property
plant and equipment in percentage of total assets, intangibles assets as percentage of total assets, earnings volatility, cash ow volatility, dividend yield
and a dividend dummy being equal to one if the stock is paying a dividend
and zero otherwise. Along with the coefcient estimates, R-squared values
and the number of observations are reported. All specications include
fund-level control variables lagged one period (fund age, fund size, expense
ratio, turnover ratio, fund capacity, returns). All specications include year
and fund xed effects. t-statistics are shown in parenthesis.

denotes 1%,

denotes 5% and

denotes 10% signicance level, respectively.
In the different regression specication in table 7, we

use the annual stock-level disposition spread measure as
the dependent variable. In Model we include the following stock characteristics as independent variables: bookto-market ratio, P/E ratio, EPS, percentage of PPE to
total assets, and the percentage of intangible assets to
total assets. As control variables, we include one year
lagged values of fund age, fund size, expense ratio, turnover ratio, fund capacity, and fund returns. We nd that
all included parameters are highly signicant at the 1%
signicance level.
In the regression specication of Model 1, we include
several nancial ratios which are commonly used for
stock valuation. The book-to-market ratio is based on a
rms total book value divided by its market capitalization and indicates whether investors think if the stock is
undervalued or overvalued (low book-to-market ratio,
high book-to-market ratio, respectively). We nd that
the book-to-market ratio is negatively linked to the level
110
S. WULFMEYER
Table 8. Disposition Effect and Stock Characteristics: Regression

Estimates
Model 1
Model 2
Model 3
Model 4
0.0364
0.0359
0.0358
0.0349
(17.61)
(17.58)
(17.57)
(17.07)
0.0001
Firm size
0.0001
(9.9)
(6.78)
0.0000
Stock turnover
0.0000
(18.54)
(11.81)
0.0006
Stock return
0.0009
(11.02)
(0.65)

0.0044
Market return
0.0178
(48.35)
(1.25)
0.0019
Stock return volatility
0.0010
(11.43)
(1.99)
Market return volatility
0.0181 0.0123
(48.7)
(3.43)
1260 593 1260 593
1260 593
N. of cases
1260 593
R-squared
0.0574
0.0758
0.0761
0.0775
Stock-level DE
This table reports the regression estimates where the disposition spread in a
given stock in a year is employed as dependent variable. The disposition
effect is rst calculated on a quarterly basis for each fund; second, aggregated to annual measures; and third, summarized for each stock in the
holdings of mutual funds during our sample period from 1980 to 2010. We
include market capitalization, rm size, turnover, stock return, return volatility, market return (S&P 500), and market return volatility as stock characteristics. Along with the coefcient estimates, R-squared values and the
number of observations are reported. All specications include fund-level
control variables lagged one period (fund age, fund size, expense ratio,
turnover ratio, fund capacity, returns). All specications include year and
fund xed effects. t-statistics are shown in parenthesis.

denotes 1%,

denotes 5% and

of the disposition effect (coefcient 0.0011 with a

t-statistics of 29.73). This means stocks with a lower
book-to-market ratio are preferably traded by less disposition-prone mutual fund managers implying that they
expect a rms management to create additional value
from given assets. This part of the regression results conrms our previous results documenting that less disposition-prone mutual fund managers tend to invest in more
expensive assets as measured by the book-to-market
ratio. Since the book-to-market ratio does not directly
provide any indication on the ability of a rm to generate
prots or cash ow for shareholders, we also include the
P/E ratio and EPS. We document a negative inuence of
the P/E ratio on the level of the disposition effect. The P/
E ratio is usually used to analyze the markets stock valuation of a rm and its shares compared to its income.
Our results indicate that less biased fund managers seem
to invest in more expensive stocks as measure by the P/E
ratio conrming our previous ndings.
We nd that more disposition-prone fund managers
tend to invest in stock with higher EPS. This indicates
that overall protability as measured by EPS is an important ratio for stock valuation of more disposition-biased
mutual fund managers. Besides the different nancial
ratios, we also analyze what kind of rms are preferred
by more disposition-affected mutual fund managers.
Variables related to the percentage of total assets invested

in PPE and in intangible assets, give an indication
whether fund managers prefer companies with rather
tangible or intangible products. We nd a positive coefcient for the PPE variable (0.0001 with a t-statistics of
5.23) and a negative relationship between the disposition
effect and the intangible assets variable (0.0003 with a
t-statistics of 6.55). Disposition-prone fund managers
prefer to invest in companies with a high level of assets
in tangible products rather than intangible assets. As
rms with a large proportion of their assets invested in
intangible assets like patents, copyrights, trademarks,
customer contracts, licensing, software, databases, etc.,
are on average more difcult to value, this result supports our hypothesis that more biased investors have difculties of valuing complex rms and are faced with a
higher level of valuation uncertainty.
In Model 2, we include several volatility-related stock
characteristics to account for stock-level uncertainty.
Results are also displayed in Table 7. The regression
specication includes earnings volatility and cash ow
volatility as stock characteristics as well as several control
variables and xed effects. Both stock variables have a
signicant effect (signicance level of 1%) on the level of
the disposition effect. The explanatory power of this
regression is lower than for Model 1, but still moderate
(R 2 of 11.06). We nd that the higher the earnings volatility, the lower the disposition effect. At a rst glance,
this result is counter-intuitive compared to our previous
ndings. When the level of uncertainty rises, the level of
the analyzed behavioral bias decreases. An explanation
might be that fund managers are more aware of behavioral biases when volatility is high. In this case, earnings
volatility can be interpreted not as a sign of valuation
uncertainty but rather as a case of attention-grabbing
stocks. In previous research, Frazzini and Lamont [2008]
document that during earnings announcements price
volatility rises and an increase in trading volume is also
evident. In their study of individual investors, the
researches focus on the attention-grabbing hypothesis,
that stocks that make news have both high turnover and
high net buying. This so-called attention hypothesis was
rst formulated by Odean [1999] and states that attention will affect the buying and selling decisions of individual investors asymmetrically. Since individuals are
reluctant to engage in short-selling and have limited abilities to evaluate a large number of stocks available in the
market, they only consider purchasing stocks which their
attention is drawn to. Our results of the earnings volatility variable can be interpreted as biased mutual fund
managers being limited in their evaluation abilities and
hence, base their trading activities more on attentiongrabbing stocks. In contrast, for the cash ow volatility
variable, we document a signicant, positive relationship

(coefcient of 0 with a t-statistics of 11.33). Fund managers investing in companies with a higher level of cash
ow volatility seem to be more disposition-prone. In line
with our previous ndings, this result contributes to the
idea that the magnitude of the observed behavioral bias
is higher when stock valuation uncertainty is higher.
In the third regression specication, we analyze how
dividends on stocks inuence the level of the disposition
effect. The results reveal that mutual fund managers with
a higher disposition level invest in dividend-paying
stocks (coefcient of 0.0004 for the variable dividend
yield and 0.0005 for the dividend-paying dummy variable with a t-statistics of 8.6 and 24.98, respectively).
Thus, mutual fund managers who are more prone to
behavioral biases seem to focus their trading activities on
dividend-paying stocks. In this case, dividend payments
are interpreted by fund managers as a sign that a companys management is feeling condent about its rms
long-term perspectives and the stocks appeal as a good
investment opportunity. Disposition-prone mutual fund
managers prefer to buy dividend-paying stocks and sell
them too early (compared to the future performance of
the respective stock). We nd that dividends can be
interpreted as an indicator for trading of attention-grabbing stocks. A previous paper by Graham and Kumar
[2006] documents that dividends may serve as a proxy
for attention-grabbing net buying behavior of individual
investors. The authors present evidence that conrms
the importance of attention for individual stock choices.
In detail, the authors document that individual investors
as a group show net buying on the day of and days following a dividend announcement, controlling for consumption or tax-motivated reasons.
In Model 4, we include all stock variables, control variables and xed effects. We conrm that the level of the
disposition effect is signicantly inuenced by stockrelated variables. We nd that book-to-market ratio, P/E
ratio, the level of intangible assets to total assets, and
earnings volatility have a signicant negative inuence
on the level of the disposition effect. Furthermore, we
document that the dividend yield and the variable dividend dummy have a pronounced positive effect on
mutual fund managers bias, conrming that not only
individual investors but also professional investors show
trading behavior motivated by attention-grabbing stocks.
To gain more insights in the relationship between the
disposition effect and different investment styles, we
include further stock-related variables. Results are displayed in Table 8. We include rm size, stock turnover,
stock return, market return, stock return volatility, and
market return volatility in our different regression specications. Model 1 to 4 of Table 8 we use the annual
111
stock-level disposition spread as the dependent variable

and include several fund-related control variables (one
year lagged values of fund age, fund size, expense ratio,
turnover ratio, fund capacity and fund returns) as well as
different xed effects. In all four model specications,
the results suggest a highly signicant positive relationship (coefcients of around 0.036 with t-statistics of
around 17.5) between the level of the disposition effect
and various stock characteristics.
In the rst regression specication in Table 8, we use
rm size and stock turnover as explanatory variables to
analyze whether size effects play a role in explaining the
level of the disposition effect. We nd that all included
parameters are signicant at the 1% signicance level.
However, all coefcients are close to zero. Firm size is
positively related to the disposition effect which means
that more biased fund managers invest in larger companies. This indicates that mutual fund managers with a
lower disposition effect try to achieve outperformance by
investing in smaller rms. We further document that
stock turnover has a negative effect on the level of the
disposition effect. Hence, fund managers with lower disposition levels prefer stock holdings which are less often
traded. High trading volume may be interpreted as a
proxy for attention-grabbing stocks which is documented for a sample of individual investors of a large
broker from 1991 to 1996 (Odean [1998]). Our results
indicate that mutual fund managers are also incline to
use trading volume as a heuristics to determine which
stocks to buy or sell.
In Model 2 of Table 8 we include stock and market
return (S&P 500) as explanatory variables. We nd that
both variables are positive and highly signicant. Stock
return has a coefcient of 0.0009 (t-statistics of 11.02)
and market return has a coefcient of 0.0178 (t-statistics
of 48.35) indicating that fund managers with a higher
disposition level invest in stocks with higher returns and
which are positively linked to the S&P 500. Mutual fund
managers are usually benchmarked against a large index
and evaluated by the information ratio (ratio of benchmark-return to benchmark-relative risk; benchmarkreturn is the expected difference between the return
earned by the fund manager and the return of an index).
A typical mutual fund manager contract is based on a
mandate to maximize the performance relative to a specic cap-weighted benchmark. Hence, mutual fund managers are incentivized to closely follow the respective
index in order to minimize the potential tracking error
compared to a cap-weighted index which supports our
ndings.
In Model 3 we include measures of stock and market
uncertainty. We document a positive and signicant
effect for stock return volatility and market volatility on
112
S. WULFMEYER
the level of the disposition effect. These results indicate

that mutual fund managers have a stronger tendency to
sell winners too early and hold on losers for too long
when valuation uncertainty is higher. Previous papers
suggest that individual investors show an increased
level of the disposition effect, when valuation uncertainty as proxied by stock volatility is higher (Kumar
[2009]).
In Model 4 we include all previously mentioned stock
characteristics. We can conrm our previous result that
stock characteristics are positively related to the level of
the disposition effect and explain parts of the behavioral
bias.
In summary, our results indicate that stock variables
explain the level of the disposition effect. We nd that
stock level determinants related to stock-level uncertainty (earnings volatility, cash ow volatility, stock
return volatility) explain the disposition effect. Our
results indicate that uncertainty at both stock and market
levels amplies mutual fund managers behavioral biases.
We also document that the trading behavior of mutual
fund managers is partly driven by trading attentiongrabbing stocks (high stock turnover stocks and
dividend-paying stocks) and by benchmark-oriented
behavior (market return). Our results suggest that the
level of the disposition effect can be explained by various
different characteristics related to fund holdings.
Uncertainty and the disposition effect

Having identied that volatility measures are positively
linked to the level of the disposition effect, we further
examine whether mutual fund managers are less prone
to realize losses when stocks are more difcult to value.
When valuation uncertainty (VU) is high, then various
mechanisms can introduce a higher level of the disposition effect. First, a prospect theory-based explanation
(Shefrin and Statman [1985]) explains why differences in
holdings drive differences in the level of the disposition
effect. Prospect theory is based on a combination of loss
aversion (Kahneman and Tversky [1979]) and stocklevel mental accounting (Thaler [1980]). Therefore, if
VU is high, higher price levels are more likely to be
observed and hence, a higher reference point for the
respective stock position is established. From a higher
level references point, the disposition effect is more likely
to be amplied.
Second, the disposition effect can be based on the
explanation of mean reversion (Odean [1998]). The
experimental empirical evidence indicates that people
show a stronger tracking behavior when VU is high
(e.g., when volatility is high) since they believe that mean
reversion is more likely.
Third, the reluctance to realize losses can also be

attributed to gambling tendencies motivating investors
to hold on to their losers when VU is high. Gamblingminded people are more likely to hold on to their positions until their stocks yield the desired extreme payoff.
Therefore, when VU is higher, the level of the disposition
effect should be higher (Kumar [2009]).
To examine whether differences in risk-taking also
exist between disposition-prone and less dispositionprone fund managers, we relate risk measures of a stock
to the fund managers level of the disposition bias and
other potentially relevant stock-related characteristics.
To analyze whether the cross-sectional variation of the
disposition effect is related to stock-level valuation
uncertainty, we sort all stock holdings based on their idiosyncratic volatility. Idiosyncratic volatility is calculated
as the variance of the residuals obtained by tting a onefactor model to the stock return time series. The idiosyncratic volatility measure for each stock is estimated each
year by using daily stock data. Annual turnover is measured as the ratio of number of shares traded and the
number of shares outstanding per month. In our regressions, we include the averages of the last year for both
variables. Firm age is calculated as the number of years
since the stock rst appears in CRSP.
First, we sort all stocks based on their average annual
idiosyncratic volatility measure during the sample period
of 19802010. Then, we calculate the mean disposition
spread in each of the ten idiosyncratic volatility decile
portfolios. As a rst indication, Figure 1a indicates that
mutual fund managers display a higher disposition effect
when valuation uncertaintyas measured by the idiosyncratic volatilityis higher, for example, when stocks
are more difcult to value. We nd that the disposition
effect is the lowest for the rst volatility decile (lowest
volatility) and the highest for the tenth idiosyncratic volatility decile. The relationship between volatility and the
disposition effect is monotonically increasing. This result
shows that the disposition effect is driven by how difcult stocks are to value. In line with this, Kumar [2009]
nds the same results for a sample of U.S. individual
investors from 1991 to 1996. Our results indicate that
mutual fund managers even though they are professionals rely on idiosyncratic volatility as a parameter for
their valuation of certain stocks. We nd that our results
are robust for stock-level and fund-level disposition
measures.
We further report sorting results for the annual disposition effect estimates for the top three (harder-to-value
stocks) and bottom three (easier-to-value stocks) idiosyncratic volatility deciles. Figure 1b presents the average
disposition effect per year. Our results indicate that the
level of the disposition bias is higher for harder-to-value
113
Figure 1. Valuation uncertainty and the Disposition Effect. Figures (a) and (b) show the mean stock-level disposition effect (PGR PLR)
for the different idiosyncratic volatility deciles. PGR is the proportion of gains realized, and it is dened as the ratio of the number of
realized winner positions and the total number of winners (realized and paper gains). PLR is the proportion of losses realized and is
dened analogously. The idiosyncratic volatility for each stock is estimated each year using daily return time series. Figure (a) displays
the average annual stock-level disposition effect for each of the ten idiosyncratic volatility deciles. Figure (b) shows the sorting results
for the annual stock-level disposition effect estimates for the top three (harder-to-value stocks) and bottom three (easier-to-value
stocks) idiosyncratic volatility deciles.
stocks. We nd that only in 2 (year 1987 and year 1990)

out of 30 years do the easier-to-value stocks show a
higher level of the disposition effect. This means that
mutual fund managers have a stronger tendency to display the analyzed behavioral bias when valuation uncertainty is high. Moreover, we document that the
disposition effect declines over time for both the harderand easier-to-value stocks. Our ndings are in line with
previous empirical ndings. For example, Kumar [2009]
also nds that the disposition effect measure for individual investors is higher when idiosyncratic volatility is
higher.
To further test our hypothesis that the level of the disposition effect is higher for harder-to-value stocks, we
estimate a panel regression with various control variables

and year-and fund-xed effects. We employ the following regression specication:
DS D a C bDS Valuation Uncertainty C bDS
k
Stock Characteristicsk C eDS
(6)
where valuation uncertainty is the idiosyncratic volatility, and stock characteristics are book-to market ratio, P/
E ratio, EPS, PPE in percentage of total assets, intangibles
assets as percentage of total assets, earnings volatility,
cash ow volatility, dividend yield, and a dividend
dummy being equal to one if the stock is paying a
114
S. WULFMEYER
dividend and zero otherwise (for the denition of these

variables please refer to Appendix B Tables B.3).
Regression results are reported in Table 9. In all
regression specications, we nd that the level of the disposition bias is higher when valuation uncertainty is
higher. We document a positive highly signicant (coefcients of 0.0015 to 0.0018 with t-statistics of 73.28 to
90.01) relationship between idiosyncratic volatility and
the disposition bias, indicating that mutual fund managers have difculties to value stocks when the level of
stock-related uncertainty is higher.
In the rst regression specication, we include a
stocks beta and the standard errors estimated from a
one-factor model as control variables. The results indicate that the relationship between the annual idiosyncratic volatility and the level of the disposition effect is
positive (coefcient of 0.0018) and signicant (t-statistics
of 89.27). Thus, mutual fund managers who are less disposition-prone tend to invest in equities with lower idiosyncratic risk. Previous papers also nd that less biased
nd managers invest in less risky equities (e.g., Ammann,
Ising, and Kessler [2012]). To control for the effects of

systematic risk on mutual fund managers behavioral
bias, we include a stocks beta estimated from a one-factor model using the S&P 500 as market index. The
regression estimates show that the disposition effect is
stronger among stocks with a higher beta (coefcient of
0.0035 and t-statistics of 40.76). Thus, fund managers
displaying a higher disposition level invest in more risky
stocks which is reected in higher beta values.
In Model 2 of Table 9, we use the same dependent
variable but consider several stock related variables as
control variables. In particular, we include the book-tomarket ratio, P/E ratio, and EPS to control for systematic
risk on the level of the observed behavioral bias among
mutual fund managers. We nd that managers with a
higher disposition bias tend to invest in growth stocks
(lower book-to-market ratio). This result is in line with
our hypothesis that more biased investors trade stocks
which are more difcult to value. In a previous paper,
Kumar [2009] documents among a sample of U.S. individual investors that investors with a higher level of the
Table 9. Disposition Effect and Valuation Uncertainty: Regression Estimates
Stock-level DE
Idiosyncratic volatility
Beta
Model 1
Model 2
Model 3
Model 4
Model 5
0.0396
(251.87)
0.0018
(89.27)
0.0035
(40.76)
0.0393
(250.21)
0.0018
(90.01)
0.0033
(38.84)
0.0001
(7.16)
0.0000
(40.6)
0.0000
(1.49)
0.0000
(1.56)
0.0002
(7.84)
0.0395
(251.92)
0.0018
(87.15)
0.0034
(39.63)
0.0392
(251.12)
0.0015
(73.28)
0.0034
(39.90)
0.0390
(250.36)
0.0016
(78.96)
0.0032
(37.93)
0.0001
(4.42)
0.0000
(39.75)
0.0000
(18.52)
0.0001
(6.91)
0.0002
(10.29)
0.0000
(21.96)
0.0000
(25.96)
0.0008
(17.89)
0.0005
(62.84)
3110 214
0.2515
Earnings-per-share
PPE
Intangible assets
Earnings volatility
Cash ow volatility
0.0000
(30.84)
0.0000
(25.99)
Dividend yield
Dividend dummy
N. of cases
R-squared
3110 254
0.2333
3110 254
0.2377
3110 254
0.2364
0.0005
(20.9)
0.0005
(61.1)
3110 214
0.2449
This table reports the results of the regression analysis of the annual stock-level disposition effect (stock-level DE) and valuation uncertainty. We measure valuation uncertainty as the idiosyncratic volatility which is calculated as the variance of the residuals obtained by tting a one-factor model to the stock return time
series. The idiosyncratic volatility measure for each stock is estimated each year by using daily stock data. In addition, we include the following stock characteristics as control variables: beta, book-to-market ratio, price-earnings ratio, earnings per share, property, plant and equipment to total assets ratio, the ratio of
intangible assets to total assets, earnings volatility, cash ow volatility, dividend yield and the dividend dummy. We further control for possible differences in
fund characteristics by including one year lagged fund-related variables: fund size, expense ratio, turnover ratio, fund capacity and lagged returns. t-statistics are
shown in parenthesis.

denotes 1%,

denotes 5% and

disposition and over-condence bias, also invest in low

book-to-market stocks. Adding the ratio of intangible
assets to total assets as an alternative variable to capture
the level of uncertainty in fundamentals, we nd a significant negative relationship (coefcient of 0.0002 with a
t-statistic of 7.84).
In Model 3, we include earnings and cash ow volatility as well as idiosyncratic volatility and beta. For robustness, we examine whether the positive relationship
between valuation uncertainty and the disposition effect
is related to other uncertainty measures. We consider
earnings and cash ow volatility as alternative measures
to capture the level of uncertainty in fundamental values.
Our results indicate that the uncertainty-disposition relation does not hold when using other fundamental based
uncertainty measures. We nd that the risk associated
with earnings and cash ows do not explain the level of
the disposition effect (coefcient of 0 with t-statistics of
30.84 and 25.99, respectively). The fundamental based
uncertainty measures indicate that the observed behavioral bias is not evident when cash ow or earnings are
more volatile.
In our fourth specication, we include control variables for dividend-paying stocks. Both control variables
related to dividends indicate that the dividend-paying
stocks contribute to a higher level of the disposition
effect. The coefcients of 0.0005 for the dividend yield as
well as the dividend dummy support our previous ndings that mutual fund managers tend to trade attentiongrabbing stocks. In this context, dividend payments serve
as an attention-grabbing event and are likely to draw
mutual fund managers attention. Similar results are
documented for individual investors (e.g., Graham and
Kumar [2006]).
In summary, we nd that the higher the level of valuation uncertainty as measured by idiosyncratic volatility,
the higher the level of the disposition bias. In addition,
we document that stocks with valuation uncertainty
related stock characteristics are generally perceived as
more difcult to value. In detail, a low book-to-market
ratio and a low ratio of intangible assets to total assets
are more traded by biased mutual fund managers. In
addition, we nd that attention-grabbing stocks (dividend-paying stocks) contribute to a higher level of the
disposition bias.
Disposition effect and betas
In the previous section, we documented that the disposition bias increases when unsystematic risk as measured
by the standard deviation of the residuals from a onefactor model raises. In this section, we analyze whether
the disposition effect is stronger when systematic risk is
higher. We investigate the relationship between fund
115
factors loading on the market factor from a one-factor

model and the level of the disposition bias.
Traditional nance is based upon the idea that higher
risk is rewarded with higher average returns. Contrary,
recent empirical evidence (Baker, Bradley, and Wurgler
[2011]) indicates that over the past 40 years U.S. portfolios comprising high-risk stocks have substantially
underperformed their lower-risk counterparts. This socalled high risk, low return puzzle offers potential
investment opportunities for mutual fund investors.
Recent empirical and theoretical evidence indicates that
low beta stocks may offer higher returns than the capital
asset pricing model (CAPM) would predict. A basic
assumption of the CAPM is that all agents invest in the
asset with the highest expected excess return per unit of
risk (Sharpe ratio). Risk preferences are adjusted by
leveraging or deleveraging this portfolio. However,
investors like mutual funds are constrained in their leverage possibilities, and they therefore have to overweight
risky assets instead of using leverage. Hence, investments
in low beta stocks may be rational and may explain why
less biased mutual fund managers invest in low beta
equities.
The persistent outperformance of low-beta stocks is
not compatible with much of nancial theory, including
the efcient market hypothesis and the CAPM. As documented in previous papers (Baker, Bradley, and Wurgler
[2011], Hong and Sraer [2012]), key principles of behavioral nance as well as structural issues including benchmarking may explain the persistence of this anomaly
across markets and across time. Theoretical behavioral
models of security prices are based on the assumption
that some market participants behave irrational. In this
context, the low-risk anomaly can be explained by a
combination of well-established biases of representativeness and overcondence leading to a demand for higherbeta stocks. Overcondence is induced by the underlying
assumption that investors agree to disagree (Hong and
Sraer [2012]). In times of high uncertainty, overcondence amplies the level of disagreement regarding
future expectations. However, the question remains why
sophisticated institutional investors like mutual fund
managers do not prot from the mistakes of the irrational crowd. One explanation may be the limits of arbitrage why the smart money does not offset the induced
pricing behavior introduced by the irrational investor
group. The performance of the asset management industry heavily depends on benchmarks which induce a
strong structural impediment. One interpretation of the
limits of arbitrage is benchmarking (Baker, Bradley, and
Wurgler [2011]). Mutual fund managers usually have a
xed-benchmark mandate (capitalization-weighted
benchmarks), which may discourage investments in low-
116
S. WULFMEYER
beta stocks. Hence, when a manager moves away from

investments in low-beta stocks, there is potential of a
higher tracking error compared to a cap-weighted index.
The trading behavior inuences the information ratio
(ratio of benchmark-return to benchmark-relative risk;
benchmark-return is the expected difference between the
return earned by the fund manager and the return of an
index), which is often the primary measure used to quantify the skills of a manager. A typical mutual fund manager contract contains an implicit or explicit mandate to
maximize the performance relative to a specic capweighted benchmark without using any leverage. Therefore, fund managers evaluated relatively to an index are
incentivized to invest in low-beta stocks.
Previous literature indicates that behavioral-based factors as well as structural characteristics of the marketplace may explain the high risk, low return anomaly.
We add to this literature by analyzing the disposition
effect as another possible driver of this anomaly. Previous literature nds that the level of the disposition effect
is inuenced by differences in investment styles. Cici
[2010] documents that the presence of the disposition
effect leads to a more value-oriented style. He documents
that more disposition-prone mutual funds are tilted
towards poor past performing stocks and that the
affected portfolios display a lower market risk exposure
(beta).
To examine whether differences in risk-taking can
explain the cross-sectional variation of the disposition
effect, we relate the beta of a stock to the level of the disposition effect and other potentially stock-related and
fund-related characteristics. Betas are estimated from a
one-factor model tted to the daily stock return series
using the S&P 500 as market index. First, we sort all
stocks based on their average annual beta during the
sample period 19802010 and divide them into deciles.
Second, we calculate the mean disposition spread in each
of the 10 beta deciles. As a rst indication, Figure 2a
indicates that mutual funds display a higher disposition
effect when beta is high. We nd that the relationship
between beta and the level of the disposition effect is positive and is almost monotonically increasing. The disposition effect is on average 3.87% in the lowest decile and
3.96% in the highest decile.
We further report sorting results for the annual disposition effect for the top three and bottom three beta deciles. Figure 2b shows the average disposition effect per
year for high- and low-beta stocks across time. Our
results indicate that fund managers with a high level of
the disposition effect tend to invest in high-beta stocks.
Only in the years 2006 and 2007 did less biased fund
managers trade stocks with a higher average beta value
compared to their more biased counterparts. This result
might be due to the nancial crisis where less biased

fund managers were also forced to sell off their losing
positions.
To further test our hypothesis that fund managers
who are more prone to the disposition effect preferably
trade high beta stocks, we estimate regressions with various control variables and xed effects. We employ the
following regression specication:
DS D a C bDS Beta C bDS
k
Stock Characteristicsk C eDS
(7)
where valuation uncertainty is the idiosyncratic volatility, and stock characteristics are book-to market ratio, P/
E ratio, EPS, PPE in percentage of total assets, intangibles
assets as percentage of total assets, earnings volatility,
cash ow volatility, dividend yield, and a dividend
dummy being equal to one if the stock is paying a dividend and zero otherwise (for the denition of these variables please refer to Appendix B Tables B.3).
Regression results including year- and fund-xed
effects are reported in Table 10. We further control
for possible differences in fund characteristics by
including the following fund-related variables such as
fund size, expense ratio, turnover ratio, fund capacity,
and returns. All control variables are lagged one year.
In all regression specications, we nd that beta has a
signicant positive inuence on the level of the disposition effect (coefcients of 0.0003 to 0.0034 with tstatistics of 3.77 to 39.86). We document a positive
relationship between the systematic risk of a stock
and the disposition bias, indicating that more biased
mutual fund managers tend to invest in stocks with
higher betas. The beta estimates of the regression
analysis conrm our graphical results of a monotonically increasing relationship between the disposition
effect and the systematic risk of traded stocks.
In Model 1 of Table 10, the only explanatory variable
that we employ is beta. We nd a positive, signicant
relationship (coefcient of 0.0003 with t-statistics of
5.76) between the systematic risk of a stock and the disposition bias while controlling for fund characteristics
and including year- and fund-xed effects.
In the second specication, we can conrm the positive relationship between the disposition bias and beta
while including book-to market ratio, P/E ratio, EPS,
and PPE in percentage of total assets as well as intangibles assets as percentage of total assets.
In the third model of Table 10 we include various additional risk measures, namely idiosyncratic volatility, earnings volatility and cash ow volatility. We nd that beta
be- comes more positive and highly signicant
117
Figure 2. Beta and the Disposition Effect.

Figures (a) and (b) show the mean stock-level disposition effect (PGR PLR) for the different beta deciles. PGR is the proportion of gains
realized, and it is dened as the ratio of the number of realized winner positions and the total number of winners (realized and paper
gains). PLR is the proportion of losses realized and is dened analogously. Betas are estimated from a one-factor model tted to the
daily stock return series using the S&P 500 as market index. Subgure (a) displays the average annual stock-level disposition effect for
each of the ten beta deciles. Subgure (b) shows the sorting results for the annual stock-level disposition effect estimates for the top
three (high-beta stocks) and bottom three (low-beta stocks) beta deciles.
(coefcient of 0.0034 with t-statistics of 41.17) compared

to the previous specications. We conrm our previous
results from section 5.2.3 documenting that the disposition effect and idiosyncratic volatility have a positive,
monotonically increasing relationship. Given that mutual
funds are not completely diversied, the higher
idiosyncratic volatility level of stock holdings of mutual

fund managers with a higher magnitude of the disposition
effect indicates that they take more risks in their portfolios. By including earnings and cash ow volatility, we
measure the inuence of valuation uncertainty. We nd a
positive effect for earnings volatility conrming previous
118
S. WULFMEYER
Table 10. Disposition Effect and Beta: Regression
Stock-level DE
Beta
Model 1
Model 2
Model 3
Model 4
Model 5
0.0388
(0.02)
0.0003
(5.76)
0.0384
(0.02)
0.0003
(5.31)
0.0000
(2.12)
0.0000
(34.72)
0.0001
(40.48)
0.0003
(31.05)
0.0002
(12.57)
0.0402
(76.8)
0.0034
(41.17)
0.0393
(0.01)
0.0002
(3.77)
0.0399
(77.04)
0.0034
(39.86)
0.0001
(6.42)
0.0000
(40.39)
0.0000
(15.66)
0.0001
(7.74)
0.0002
(9.34)
0.0015
(75.73)
0.0000
(21.28)
0.0000
(24.04)
0.0007
(16.14)
0.0006
(67.38)
3110 237
0.2799
Earnings-per-share
PPE
Intangible asstes
Earnings volatility
Cash ow volatility
0.0017
(84.95)
0.0000
(30.07)
0.0000
(24.32)
Dividend yield
Dividend dummy
N. of cases
R-squared
4920 274
0.1952
4920 274
0.2016
3110 237
0.2641
0.0008
(38.36)
0.0005
(79.74)
4920 032
0.2102
Estimates This table reports the results of the regression analysis of the annual stock-level disposition effect (stock-level DE) and betas estimated from a one-factor
model tted to the daily stock return series using the S&P 500 as market index. We include year- and fund-xed effects as control variables. We further control
for possible differences in fund characteristics by including the following fund-related variables: fund size, expense ratio, turnover ratio, fund capacity, and
returns. All variables are lagged one year. t-statistics are shown in parenthesis.

denotes 1%,

denotes 5% and

ndings by Kumar [2009] documenting that if fundamentals-based uncertainty is high, the disposition effect is
stronger.
In Model 4, we include the two dividend-related
measures. We document a positive beta as well as a positive relationship between the dividend measures and the
disposition bias. This means that more disposition-prone
fund managers invest in equities with a higher dividend
yield. When interpreting dividends as a signaling feature
of stocks, our results indicate that not only individual
investors but also professional investorslike mutual
fund managersare prone to trade attention-grabbing
stocks. A previous paper by Graham and Kumar [2006]
shows that dividends can be interpreted as a proxy for
attention-grabbing trading behavior of individual investors. Our results are in line with previous papers suggesting that attention-grabbing stocks amplify the
disposition effect.
In Model 5 of Table 10, we show that beta is also positive and highly signicant (coefcient of 0.0034 with tstatistics of 39.86) including all stock-related variables.
Our results indicate that mutual fund managers with a
lower level of the observed behavioral bias tend to invest
in less risky assets. All stock-related variables are signicant when included in regression model 5. We can
conrm the positive relationship among the book-tomarket ratio, the P/E ratio, and the two dividend-related
measures.
In summary, our results show that less dispositionprone investors tend to invest in less risky equities.
Using graphical and regression-based approaches, we
nd a monotonically increasing relationship between
the level of the disposition effect and stock betas.
Ours results indicate that mutual fund managers who
are more prone to behavioral biases tend to hold
high-beta stocks in their portfolio. Investing in high
beta stocks which do not have as high returns as predicted by the CAPM, may be one reason why more
biased fund managers underperform compared to
their less biased counterparts.
Summary and conclusion

As professional investors, mutual funds are assumed
to show rational behavior and by searching for mispriced securities and market anomalies are supposed
to make nancial markets more efcient. In this
paper, we show that the average fund is dispositionprone, a behavioral bias which is usually attributed to
individual investors.
On a large sample of actively managed U.S. mutual

funds trading decisions from 1980 to 2010, this paper
nds robust empirical evidence for the existence of the
disposition effect. First, we document that, on average,
mutual fund managers in our sample are more likely to
realize capital gains than losses. Second, we demonstrate
that characteristics of mutual funds holdings can play a
fundamental role in impacting the magnitude of the disposition effect. Stock characteristics are found to signicantly affect the disposition bias. In particular, trading
stocks with a high level of valuation uncertainty signicantly increases the likelihood and extent of the disposition bias in mutual fund managers investment decisions.
Third, we show that the level of systematic risk (i.e.,
beta) is monotonically increasing with the level of the
disposition effect. We nd that fund managers with a
high-level disposition effect tend to invest in high beta
stocks. As high-beta stocks empirically do not outperform low-beta stocks (Baker, Bradley, and Wurgler
[2011]), our results contribute to explain the underperformance of biased mutual funds. Fourth, we document
that the disposition effect is partly driven by trading of
attention-grabbing stocks. We nd that mutual fund
managers tend to invest in dividend-paying stocks where
the dividend payment serves as a signaling feature of a
rm. In addition, investments in smaller rms with a
lower turnover also contribute to avoid behavioral biases
in organizational decision making. The results suggest
that a signicant de-biasing of mutual funds outcomes
can be achieved through taking not only stock-related
performance measure but also additional variables into
account when making trading decisions. The ndings
call for more research in the area of the relationship
between trading decisions and behavioral biases.
Our results shed light on the interaction of stock
holdings characteristics and behavioral biases. The
behavior of mutual fund managers is consistent with
recent evidence on the disposition effect in a professional setting presented by Scherbina and Jin [2011].
We add to the research stream on mutual funds and
their decision-making processes. Moreover, we contribute to the ongoing discussions of the drivers of
behavioral biases. Our ndings suggest that fund
managers display difculties in evaluating stocks and
are also prone to employing heuristics in making
their trading decisions.
Our ndings do not come without limitations.
First, our investigation is limited to one specic type
of decision-making bias, namely the disposition effect.
Second, we are looking at one particular setting
(equity mutual funds) during a specic time period
(19802010). Furthermore, our analysis is limited to
exploring how stock characteristics affect the
119
magnitude of the disposition bias. Future research

may further analyze the circumstances and organizational settings enhancing or decreasing behavioral
biases like the disposition bias.
Notes
1. Source: 2012 Investment Company Fact Book, Investment
Company Institute.
2. See Jensen [1986], Malkiel [1995], Daniel, Grinblatt, Titman, and Wermers [1997].
3. Details on the sample selection procedure can be found in
Appendix A.
4. The share weighted average of all reported purchases is
taken as the weighted average purchase price.
5. Previous papers (e.g., Cici [2010]) analyzed differences in
the assumptions about when a trade takes place (at the
beginning, during or at the end of a quarter). The results
indicate that variations in assumptions about the timing
of a trade do not change the results with respect to the
presence of the disposition effect.
6. Nevertheless, the majority of mutual funds reported their
holdings on a quarterly basis to Thomson prior to June
2005
References
Alevy, J. E., M. Haigh and J. List. Information Cascades: Evidence from a Field Experiment with Financial Market Professionals. Journal of Finance, 62, (2007), pp. 151180.
Ammann, M., A. Ising and S. Kessler. Disposition Effect and
Mutual Fund Performance. Applied Financial Economics,
22, (2012), pp. 119.
Baker, M., B. Bradley and J. Wurgler. Benchmarks as Limits to
Arbitrage: Understanding the Low-Volatility Anomaly.
Financial Analyst Journal, 67, (2011), pp. 4054.
Barber, B. and T. Odean. All that Glitters: The Effect of Attention and News on the Buying Behavior of Individual and
Institutional Investors. Review of Financial Studies, 21,
(2008), pp. 785818.
Ben-David, I. and D. Hirshleifer. Are Investors Really Reluctant to Realize Their Losses? Trading Responses to Past
Returns and the Disposition Effect. Review of Financial
Studies, 25, (2012), pp. 24852532.
Busse, J. A. and Q. Tong. Mutual Fund Industry Selection and
Persistence. Review of Asset Pricing Studies, 2, (2012), pp.
245274.
Caballero, R. and A. Krishnamurthy. Collective Risk Management in a Flight to Quality Episode. Journal of Finance, 53,
(2008), pp. 21952230.
Chiang, M.-H. and H.-Y. Huang. Do Mutual Fund Flows
Drive the Disposition Behavior of Fund Managers? Working Paper available at SSRN 1571406 (2010).
Cici, G. The Relation of the Disposition Effect to Mutual Fund
Trades and Performance. Working Paper available at
SSRN 645841 (2010).
Cici, G. The Prevalence of the Disposition Effect in Mutual
Funds Trades. Journal of Financial and Quantitative Analysis, 47, (2012), pp. 795820.
Coval, J. D. and T. Shumway. Do Behavioral Biases Affect
Prices? Journal of Finance, 1, (2005), pp. 134.
120
S. WULFMEYER
Daniel, K., M. Grinblatt, S. Titman and R. Wermers. Measuring Mutual Fund Performance with Characteristic-Based
Benchmarks. Journal of Finance, 52, (1997), pp. 1035
1058.
Daniel, K., D. Hirshleifer and A. Subrahmanyam. Investor
Psychology and Security Market Under- and Overreactions. Journal of Finance, 53, (1998), pp. 18391886.
Dhar, R. and N. Zhu. Up Close and Personal: Investor Sophistication and the Disposition Effect. Management Science,
52, (2006), pp. 726740.
Dorn, D. and G. Strobl. Rational Disposition Effects: Theory
and Evidence. Drexel University, Working Paper (2011).
Eshraghi, A. and R. Tafer. Fund Manager Overcondence
and Investment Performance: Evidence from Mutual
Funds. Working Paper available at SSRN 2146864 (2012).
Evans, R. Mutual Fund Incubation. Journal of Finance, 65,
(2010), pp. 15811611.
Frazzini, A. The Disposition Effect and Underreaction to
News. Journal of Finance, 61, (2006), pp. 20172046.
Frazzini, A. and A. Lamont. Dumb Money: Mutual Fund
Flows and the Cross-Section of Stock Returns. Journal of
Financial Economics, 88, (2008), pp. 299322.
Genesove, D. and C. Mayer. Loss Aversion and Seller Behavior: Evidence from the Housing Market. Quarterly Journal
of Economics, 116, (2001), pp. 12331260.
Gil-Bazo, J. and P. Ruiz-Verdu. The Relation Between Price
and Performance in the Mutual Fund Industry. Journal of
Finance, 64, (2009), pp. 21532183.
Goetzmann, W. and M. Massa. Disposition Matters: Volume,
Volatility and Price Impact of a Behavioral Bias. Journal of
Portfolio Management, 34, (2008), pp. 103125.
Graham, J. R. and A. Kumar. Do Dividend Clienteles Exist?
Evidence on Dividend Preferences of Retail Investors.
Journal of Finance, 61, (2006), pp. 13051336.
Grinblatt, M. and M. Keloharju. What Makes Investors
Trade? Journal of Finance, 56, (2001), pp. 589616.
Hartzmark, S. and D. H. Solomon. Efciency and the Disposition Effect in NFL Prediction Markets. Quarterly Journal
of Finance, 2, (2012), pp. 125200.
Heath, C., S. Huddart and M. Lang. Psychological Factors and
Stock Option Exercise. Quarterly Journal of Economics,
114, (1999), pp. 601627.
Hirshleifer, D. Investor Psychology and Asset Pricing. Journal of Finance, 56, (2001), pp. 15331597.
Hong, H. G. and D. A. Sraer. Speculative Betas. Journal of
Finance, (Forthcoming).
Huang, J., K. Wei and H. Yan. Participation Costs and the
Sensitivity of Fund Flows to Past Performance. Journal of
Finance, 62, (2007), pp. 12731311.
Huddart, S. and V. Narayanan. An Empirical Examination of
Tax Factors and Mutual Funds Stock Sales Decision.
Review of Accounting Studies, 7, (2002), pp. 319341.
Jensen, M. The Performance of Mutual Funds in the Period
194564. Journal of Finance, 23, (1986), pp. 389416.
Kacperczyk, M., C. Sialm and L. Zheng. Unobserved Actions
of Mutual Funds. Review of Financial Studies, 21, (2008),
pp. 23792416.
Kahneman, D. A Psychological Perspective on Economics.
American Economic Review, 93, (2003), pp. 162168.
Kahneman, D. and A. Tversky. Prospect Theory: An Analysis
of Decision Under Risk. Econometrica, 46, (1979),
pp. 171185.
Kaustia, M., E. Alho and V. Puttonen. How Much Does

Expertise Reduce Behavioral Biases? The Case of Anchoring
Effects in Stock Return Estimates. Financial Management,
37, (2008), pp. 391411.
Kumar, A. Hard-To-Value Stocks, Behavioral Biases, and
Informed Trading. Journal of Financial and Quantitative
Analysis, 44, (2009), pp. 13751401.
Kumar, A. and S. Lim. How Do Decision Frames Inuence
the Stock Investment Choices of Individual Investors?
Management Science, 54, (2008), pp. 10521064.
Locke, P. and S. Mann. Do Professional Traders Exhibit Loss
Realization Aversion? Texas Christian University, Working Paper (2000).
Malkiel, B. Returns from Investing in Equity Mutual
Funds 1971 to 1991. Journal of Finance, 50, (1995),
pp. 549572.
OConnell, P. G. and M. Teo. Institutional Investors, Past
Performance and Dynamic Loss Aversion. Journal of Financial and Quantitative Analysis, 44, (2009), pp. 155188.
Odean, T. Are Investors Reluctant to Realize Their Losses?
Journal of Finance, 53, (1998), pp. 17751798.
Odean, T. Do Investors Trade Too Much? American Economic Review, 89, (1999), pp. 12791298.
Pool, V. K., N. Stoffman and S. E. Yonker. No Place Like
Home: Familiarity in Mutual Fund Manager Portfolio
Choice. Review of Financial Studies, 25, (2012), pp. 2564
2599.
Puetz, A. and S. Ruenzi. Overcondence Among Professional
Investors: Evidence from Mutual Fund Managers. Journal
of Finance and Accounting, 38, (2011), pp. 684712.
Ringov, D. Can Organizations Mitigate Individual Biases?
Evidence from Mutual Fund Investment Decisions.
DRUID Society Conference, June 1517, DRUID (2012).
Rosa, R. D. S., H. M. To and T. S. Walter. Tests of the Disposition Effect among UK Managed Funds. Working Paper,
University of Western Australia (2005).
Scherbina, A. and L. Jin. Inheriting Losers. Review of Financial Studies, 24, (2011), pp. 786820.
Seru, A., T. Shumway and N. Stoffman. Learning by Trading.
Review of Financial Studies, 23, (2010), pp. 705739.
Shapira, Z. and I. Venezia. Patterns of Behavior of Professionally Managed and Independent Investors. Journal of
Banking and Finance, 40, (2001), pp. 15731587.
Shefrin, H. and M. Statman. The Disposition to Sell Winners
Too Early and Ride Losers Too Long: Theory and Evidence. Journal of Finance, 40, (1985), pp. 777790.
Singal, V. and Z. Xu. Selling Winners, Holding Losers: Effect
on Fund Flows and Survival of Disposition-Prone Mutual
Funds. Journal of Banking and Finance, 35, (2011), pp.
27042718.
Solomon, D. H., E. F. Soltes and D. Sosyura. Winners in the
Spotlight: Media Coverage of Fund Holdings as a Driver of
Flows. Journal of Financial Economics, 113, (2014), pp. 53
72.
Thaler, R. Toward a Positive Theory of Consumer Choice.
Journal of Economic Behavior and Organization, 1, (1980),
pp. 3960.
Thaler, R. Mental Accounting and Consumer Choice. Marketing Science, 4, (1985), pp. 199214.
Weber, M. and C. F. Camerer. The Disposition Effect in Securities Trading: An Experimental Analysis. Journal of Economic Behavior and Organization, 33, (1998), pp. 167184.
Wermers, R. Mutual Fund Performance: An Empirical

Decomposition into Stock-Picking Talent, Style, Transactions Costs, and Expenses. Journal of Finance, 55, (2000),
pp. 16551695.
Wermers, R. R. Is Money Really Smart? New Evidence on
the Relation Between Mutual Fund Flows, Manager Behavior, and Performance Persistence. Working Paper, University of Maryland, College Park (2003).
Appendix
Appendix A: Mutual fund sample selection

This appendix provides additional details of how we constructed our data. The data in this paper are collected
from several sources. We start with a sample of all
mutual funds in the CRSP Survivor-Bias-Free U.S.
Mutual Fund Database (CRSP MF, henceforth) covering
the period 19802010. Both databases are provided by
Wharton Research Data Services (WRDS). The focus of
our analysis is on domestic equity mutual funds for
which the holdings data are most complete and reliable.
We start constructing our sample with the universe of
all open-end funds listed by the Survivor-Bias-Free U.S.
Mutual Fund Database maintained between January
1980 and December 2010, inclusive. The database covers
all (live and dead) equity, bond, and money market
mutual funds (MFs) since December 1961. CRSP MF
database provides a complete historical record for each
fund, including fund name, identifying information (e.g.,
fund number, fund name), start and end dates, net asset
values, loads, various classication system for investment
category, assets under management, returns, fund families, and further items. The initial sample is downloaded
from CRSP MF database and the sample constitutes of
1,193,818 observations and 49,004 funds.
Since the estimation of mutual funds decision-making process (here: disposition effect) also requires holding-level data on fund portfolio decisions, we use a
second data source, namely the Thomson Reuters
Mutual Fund Holdings database (TR, henceforce). The
database contains survivor-bias-free data on quarter-end
holdings which are reported by U.S.-based mutual funds
in the mandatory Securities and Exchange Commission
(SEC) lings. Mutual funds in the United States are
required by the SEC to report their portfolio holdings
semi-annually prior to June 2005 and since then changed
to a quarterly reporting mode6.
The main dataset is created by merging the CRSP with
the TR MF database by using the MFLinks document,
also provided by WRDS. We obtain fund stock holdings
from Thomsons SP12 database since 1980, which in
turn determines the starting date of our analysis. Thomson sometimes backlls gaps with information from previous quarters which is identied by the variable rdate
121
(reporting date). Besides the quarterly frequency of holding reports, a further limitation is that short positions
are unobserved. Also, assumptions about holding returns
and trade timing have to be made. In addition, fdate is
reported referring to the actual date for which the holdings are valid. We follow standard practice and limit our
sample of holdings to those observations where the fdate
is equal or larger than the rdate to avoid the use of stale
data in our analysis (Pool, Stoffman, and Yonker [2012]).
Since we are interested in the domestic portion of
funds portfolios, we remove holdings in rms headquartered outside the United States. We further limit our
analysis to actively-managed equity funds and thereby
exclude index funds, international funds and funds
focused on bonds, governments, real estate investment
trusts, convertible debt, precious metals, and other asset
classes as these types of funds generally hold and trade in
very small quantities of domestic equity. In detail, during
each quarter, we include only mutual funds having a
self-declared investment objective of aggressive growth,
growth, and growth and income, income, at the beginning of each quarter. As CRSP MF provides one observation per period for each share class of each mutual fund
we use the unique wcn (Wharton Financial Institution
Center Number) fund number for aggregation of fund
data across share classes into one observation per fundyear. We calculate weighted-averages using the total net
assets of each class as weight for characteristics that vary
across fund share classes such as returns and expense
ratios. For the total net assets of a fund, we measure the
sum of the total net assets of all the classes of that fund.
We merge the TR database with the CRSP MF database using WRDSs MFLinks, a table which links Thomsons fund identier with those of the CRSP MF
database. Approximately 92% of the target universe is
matched. The unlinked U.S. equity funds are mainly
small, defunct funds where accurate information for a
proper linking procedure is not available. In addition,
fairly new funds are also less likely to be linked since
they are not yet documented in the TR MF database.
We base the selection of our sample on varies ltering
methods which are also applied in previous similar studies (e.g., Kacperczyk, Sialm, and Zheng [2008], Pool et al.
[2012]). We eliminate all balanced, bond, money market,
sector and international funds, as well as funds which
are not primarily invested in equities. In detail, we use
the classication information from Lipper, Strategic
Insight, Wiesenberger Objective and the variable policy.
The following Lipper classication codes are used to
determine the funds as equity: LCCE, LCGE, LCVE,
MLCE, MLGE, MLVE, SCCE, SCGE, SCVE, MCCE,
MCGE, or MCVE. Further, funds are also dened as
equity if they have AGG, GMC, GRI, GRO, ING, or SCG
122
S. WULFMEYER
as Strategic Insight classication code, or GRO, LTG,

MCG, or SCG as Wiesenberger Objective code. Finally, if
the fund has a CS policy (common stocks are the securities mainly held by the fund), we also classify the fund as
equity fund. We eliminate all funds which do not meet
the above mentioned criteria.
In order to address the problem of the incubation bias
(see Evans [2010], Pool et al. [2012]), we drop all observations where the month for the observation is prior to
the reported fund starting month in CRSP MF database.
In addition, we also exclude observations in CRSP MF
where the fund had less than $5 million under management or where fewer than 11 stock holdings are identied (Pool et al. [2012], Scherbina and Jin [2011]) in the
previous month since incubated funds tend to be smaller.
The rationale behind is to keep our analysis away from
errors in the database and underreporting issues.
The resulting sample is merged with the TR data
using MFLinks. The MFLinks tables provide information
in order to combine the CRSP MF database that covers
mutual fund returns, loads/fees/expenses and related
information to equity holdings data in the TR MF datasets. In a rst step, we obtain the Wharton Financial
Institution Center Number (wcn) for each share class
in CRSP MF database and investment objective codes
(ioc) from the TR database. In a second step, the wcn is
used to nd the associated fundno and date range in the
TR data bank. Funds without a record in MFLinks are
dropped from the sample.
Even after the previous mentioned ltering, our sample still contains a number of nonequity and international funds. Therefore, we apply further ltering
mechanisms (based on Kacperczyk, Sialm, and Zheng
[2008]) in order to ensure a sample consisting only of
U.S.-based equity funds. We rst look at the percentage
invested in common shares (per com) from the CRSP
MF annual summary le and exclude all funds that on

average hold less than 80% or more than 105% in common stocks. In order to check the robustness of the
MFLinks merger, we compare total fund assets (tna) in
CRSP MF to tna in the TR database.
Absdiff D
TNACRSP TNATR
TNACRSP
(8)
If the median absolute difference for a particular fund

over all overlapping data observations is larger than 1.3 or
smaller than 1/1.3, we also drop the fund from our sample.
In order to focus on actively managed funds, we
further exclude index funds and exchange-traded
funds (ETFs). In a rst step, we eliminate all funds
which are marked as index funds, ETFs, or exchangetraded notes (ETNs) in the CRSP MF database. We
also apply a second lter procedure based on the
works of Gil-Bazo and Ruiz-Verdu [2009] and Pool
et al. [2012]. To do so, we use fund names and drop
any fund name including any of the following strings:
Index, Idx, Ix, Indx, Nasdaq, Dow, Mkt,
DJ, S&P 500, Barra, DFA, Vanguard, ETF,
SPDR, ETN, Powershares, WisdomTree,
Tracker, and Profunds.
Appendix B: Variable Denition
The tables in Appendix B briey outline how our
main variables used in the empirical analysis are constructed. Data come from three different data sources
used in this paper: quarterly fund holdings data from
Thomson Financial, quarterly mutual fund characteristics from Center for Research in Security Prices
(CRSP), and daily (monthly) stock return les from
CRSP.
123
Table B.1. Mutual Fund Data.

Variable name
Disposition effect
Fund performance
Portfolio volatility
TNA
Trades
Turnover ratio
Volume
Description and denition

Disposition effect calculated according to Odean (1998)s method in a particular quarter. It is calculated as PGR (proportion
gains realized) minus PLR (proportion loses realized).
Quarterly market-adjusted average returns of funds total quarterly return, i.e. the return on the funds portfolio, including
reinvested dividends
Portfolio volatility Mutual funds realized monthly (quarterly) portfolio volatility calculated based on the returns of
portfolios
Total net assets in a particular quarter. Reported in $ millions.
Number of all executed transactions in a particular quarter
Average of the absolute values of all purchases and sales in a particular quarter divided by the average of the portfolio
values at the beginning and end of a particular quarter. It is the minimum of aggregated sales or aggregated purchases
of securities, divided by the average 12-month Total Net Assets of the fund. The ratio is expressed as a percentage of
the fund.
Sum of the absolute values of all purchases and sales in a particular quarter.
Table B.2. Mutual Fund Characteristics.

Variable name
12b-1 fees
Expense ratio
Front load
Fund age
Fund costs
Fund size
Funds capacity
Funds trading strategy
Managers experience
New manager
Rear load
Team-managed

Is reported as the ratio of the total assets attributed to marketing and distribution costs. Represents the actual fee paid in
the most recently completed scal year as reported in the Annual Report Statement of Operations.
Ratio of total investment that shareholders pay for the funds operating expenses, which include 12b-1 fees. It may include
waivers and reimbursements, causing it to appear to be less than the fund management fees.
Front load for investments represents maximum sales charge at breakpoint.
Fund age in years, relative to the date the fund was rst offered
Measuring the percent of a funds assets paid by shareholders to cover the funds operating expenses, including
management fees.
Fund size is measured as quarterly total net asset value in million $
Dummy variable equal to 1 if the fund is open to new investors; 0 otherwise.
Dummy variable for a funds investment objective code (ioc) based on Thomson Financial database
Years of manager at specic fund.
Dummy variable that equals one if the fund manager has been at the fund for less than three years.
Fee which is charged when withdrawing fun
Dummy variable that equals one if the fund is managed by
Table B.3. Stock Characteristics

Variable name
Cash ow volatility
Dividend paying dummy
Dividend yield
Earnings-per-share
Earnings volatility
Firm age
Firm size
Intangible assets
Market capitalization
Market return
Market return volatility
Past 12-month stock return
PPE
Stock beta
Stock price
Stock return volatility
Stock turnover
Data source
Time period
Is dened as book value divided by the current market price.

Is the standard deviation of cash ows. Cash ows are dened as
income before extraordinary items plus depreciation and
amortization.
Dummy variable which is equal to 1 if the stock is paying a dividend
and 0 Otherwise.
Is calculated as annualized dividend rate divided by monthly closing
price. Reported in percentage.
Net income divided by common shares outstanding.
Is the standard deviation of earnings.
Is the number of years since the stock rst appears in the CRSP
database.
Is dened as the natural log of market capitalization (in $ million), which
is calculated as the number of shares outstanding (shrout) times
price.
Is the volatility of the residuals obtained by tting a 1-factor model to
the daily stock price time series. The idiosyncratic volatility for each
stock is estimated each year by using daily returns.
Ratio of intangible assets to total assets.
Shares outstanding times price. The market value of shares traded is
calculated as daily closing price minus daily trading volume.
Return on the S&P 500.
Is the standard deviation of daily returns of the S&P 500.
Stock returns of the last year.
Ratio of property, plant and equipment to total assets.
Is the stock price divided by earnings (net income).
Is the beta coefcient obtained from tting a one-factor model to the
daily stock price time series.
Is the daily stock price.
Is the standard deviation of daily stock returns.
Is the ratio of the number of shares traded and shares outstanding.
COMPUSTAT
COMPUSTAT
Monthly
Monthly
CRSP
Daily
COMPUSTAT
Monthly
COMPUSTAT
COMPUSTAT
CRSP
Monthly
Monthly
Daily
CRSP
Daily
CRSP
Daily
COMPUSTAT
COMPUSTAT
Monthly
Monthly
CRSP
CRSP
CRSP
COMPUSTAT
COMPUSTAT
CRSP
Daily
Daily
Daily
Monthly
Monthly
Daily
CRSP
CRSP
CRSP
Daily
Daily
Daily

Wulf Meyer 2016

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Wulf Meyer 2016

Hochgeladen von

Copyright:

Verfügbare Formate

Journal of Behavioral Finance

ISSN: 1542-7560 (Print) 1542-7579 (Online) Journal homepage: http://www.tandfonline.com/loi/hbhf20

Irrational Mutual Fund Managers: Explaining

Published online: 25 May 2016.

Submit your article to this journal

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

Date: 02 June 2016, At: 02:32

JOURNAL OF BEHAVIORAL FINANCE

Irrational Mutual Fund Managers: Explaining Differences in Their Behavior

Downloaded by [University of Cambridge] at 02:32 02 June 2016

University of St. Gallen

Mutual funds; Disposition

CONTACT Sina Wulfmeyer

Downloaded by [University of Cambridge] at 02:32 02 June 2016

between the magnitude of the stock-level disposition

book per price value. The ratio of intangible assets to

Downloaded by [University of Cambridge] at 02:32 02 June 2016

JOURNAL OF BEHAVIORAL FINANCE

[2010]). For example, Wermerss [2003] ndings suggest

biases. In order to understand which factors support or

Downloaded by [University of Cambridge] at 02:32 02 June 2016

gains and losses. They divide different types of gambles

loser funds and investors who systematically add loser

Downloaded by [University of Cambridge] at 02:32 02 June 2016

JOURNAL OF BEHAVIORAL FINANCE

To arrive at the nal sample used in the empirical

contrarian strategy or trend-following behavior. Past

Table 1. Summary Statistics of Fund Characteristics. This table

2260 535 12.54 9.9363

Downloaded by [University of Cambridge] at 02:32 02 June 2016

sample period from rst-quarter 1980 to last-quarter

Table 3. Statistics for Value of Trades

Downloaded by [University of Cambridge] at 02:32 02 June 2016

JOURNAL OF BEHAVIORAL FINANCE

where SharePrc is the respective share price, SharesBuy

If the disposition spread is positive, then the respective

% of funds Mean Std. Dev p25

Downloaded by [University of Cambridge] at 02:32 02 June 2016

calculated for each single fund separately. Second, we

preference for selling a stock by virtue of having a gain

JOURNAL OF BEHAVIORAL FINANCE

Downloaded by [University of Cambridge] at 02:32 02 June 2016

Disposition effect and stock characteristics

Table 5. Summary Statistics of Stock

Mean Std. Dev

CharacteristicsThis table presents the characteristics for the 15,110 different

In Table 6 we report the summary statistics for the

Downloaded by [University of Cambridge] at 02:32 02 June 2016

lower disposition deciles and decreases monotonically

behavioral biases are stronger. Hence, his empirical

Downloaded by [University of Cambridge] at 02:32 02 June 2016

JOURNAL OF BEHAVIORAL FINANCE

general, a higher P/E ratio suggests that the market

where stock characteristics are book-to-market ratio, P/E

Table 7. Disposition Effect and Accounting-Related Stock Characteristics: Regression Estimates

In the different regression specication in table 7, we

Table 8. Disposition Effect and Stock Characteristics: Regression

Downloaded by [University of Cambridge] at 02:32 02 June 2016

of the disposition effect (coefcient 0.0011 with a

Variables related to the percentage of total assets invested

Downloaded by [University of Cambridge] at 02:32 02 June 2016

JOURNAL OF BEHAVIORAL FINANCE