THE JOURNAL OF FINANCE • VOL. LXIX, NO.

4 • AUGUST 2014

Time-Varying Fund Manager Skill

MARCIN KACPERCZYK, STIJN VAN NIEUWERBURGH, and LAURA VELDKAMP∗

ABSTRACT
We propose a new definition of skill as general cognitive ability to pick stocks or time
the market. We find evidence for stock picking in booms and market timing in reces-
sions. Moreover, the same fund managers that pick stocks well in expansions also time
the market well in recessions. These fund managers significantly outperform other
funds and passive benchmarks. Our results suggest a new measure of managerial
ability that weighs a fund’s market timing more in recessions and stock picking more
in booms. The measure displays more persistence than either market timing or stock
picking alone and predicts fund performance.

A LARGE LITERATURE STUDIES whether investment managers add value for their
clients, and if so how. One way to shed light on these questions is to decompose
fund performance into stock picking and market timing. Previous work esti-
mates picking and timing implicitly assuming that each manager is endowed
with a fixed amount of each skill. But stock picking and market timing are not
talents one is born with. They are the result of time spent analyzing data. Like
workers in other jobs, fund managers may choose to focus on different tasks at
different points in time. This simple idea leads us to evaluate fund manager
skill in a way that allows its nature to change, depending on economic con-
ditions. Our results show that successful managers pick stocks well in booms
and time the market well in recessions. This suggests that stock picking and
market timing are tasks, rather than distinct and permanent talents. Skilled
managers can successfully perform these tasks, but how much of each they
choose to do depends on the market environment. As financial blog ZeroHedge
writes, “It is hard for a portfolio manager to focus on the nuances of stock se-
lection when the prospects of a U.S. recession keep rising. . . . Simply put, the
macro is overwhelming the micro.”1
Understanding exactly how managers add value for their clients is important
because a large and growing fraction of individual investors delegate their port-
folio management to professional investment managers.2 Yet a significant body

∗ Marcin Kacperczyk is with Imperial College and NBER. Stijn van Nieuwerburgh and Laura
Veldkamp are with NYU Stern School of Business, NBER, and CEPR. We thank an anonymous
referee, the Editor (Cam Harvey), and an Associate Editor for many useful suggestions. We thank
the Q-group for their generous financial support.
1 Published September 25, 2011.
2 In 1980, 48% of U.S. equity was directly held by individuals—as opposed to being held through

intermediaries; by 2007, that fraction was down to 21.5% (French (2008, table 1)). At the end of
DOI: 10.1111/jofi.12084

1455
1456 The Journal of FinanceR

of evidence finds that the average actively managed fund does not outperform
passive investment strategies, net of fees and after controlling for differences in
systematic risk exposure. Instead, a small subset of funds persistently outper-
form.3 The consensus view from this literature is that there is some evidence of
stock-picking ability among the best managers, but little evidence for market
timing.4 One reason these studies fail to detect market timing is because it
is typically displayed only in recessions, which account for a small fraction of
the sample periods studied. Our approach differs from the typical approach
in the literature, which studies stock picking and market timing in isolation,
unconditional on the state of the economy. Once we condition on the state of
the economy, we find a surprising result: skilled managers successfully perform
both tasks. Those who are good stock-pickers in booms are also good market-
timers in recessions. This result not only holds for the standard National Bu-
reau of Economic Research (NBER) recession indicator, but also for measures
of aggregate economic activity that are available in a more timely fashion.
The fact that only a subset of managers add value makes identification of
these skilled managers important. Therefore, a second contribution of the paper
is to develop a new real-time measure for detecting managerial skill, one that
gives more weight to a fund manager’s market-timing success in recessions and
stock-picking success in booms. This new measure predicts performance and
displays persistence of up to one year.
To measure skill, we construct estimates of stock picking (the product of a
fund’s portfolio weights in deviation from market weights and the firm-specific
component of stock returns) and market timing (the product of portfolio weights
in deviation from market weights and the aggregate component of stock re-
turns) for each firm. We regress these timing and picking variables on a reces-
sion indicator to determine if the nature of skill changes significantly over the
business cycle. We find that the average fund manager exhibits better stock
picking in booms and better market timing in recessions. Moreover, results
from quantile regressions show that it is the most skilled managers that vary
the use of their skills most over the business cycle.
To show that skilled managers exist, we select the top 25% of funds in terms
of their stock-picking ability in expansions and show that the same group has

2008, $9.6 trillion was invested with such intermediaries in the United States. Of all investment
in domestic equity mutual funds, about 85% was actively managed (2009 Investment Company
Factbook).
3 See, for example, Pástor and Stambaugh (2002), Kacperczyk, Sialm, and Zheng (2005, 2008),

Kacperczyk and Seru (2007), Christoffersen, Keim, and Musto (2007), Cremers and Petajisto
(2009), Koijen (2014), Baker et al. (2010), Huang, Sialm, and Zhang (2011), Amihud and Goyenko
(2013), and Cohen, Polk, and Silli (2011).
4 See, for example, Graham and Harvey (1996), Ferson and Schadt (1996), Daniel et al. (1997),

Becker et al. (1999), and Kacperczyk and Seru (2007). Notable exceptions are Mamaysky, Spiegel,
and Zhang (2008), who find evidence for market timing using Kalman filtering techniques; Bollen
and Busse (2001) and Elton, Gruber, and Blake (2012), who find evidence of market timing using
higher frequency holdings data; and Ferson and Qian (2004), who look at market timing in different
economic conditions. Avramov and Wermers (2006) find that active management is valuable in an
environment where both benchmark returns and managerial skill are predictable.
 Time-Varying Fund Manager Skill 1457

significant market-timing ability in recessions; the remaining funds show no

such ability. Conversely, we can select the top 25% of funds in terms of their
market-timing ability in recessions and show that this same group has sig-
nificant stock-picking ability in booms. These top funds produce unconditional
fund returns that are 50 to 80 basis points per year in excess of the other funds,
before expenses and on a risk-adjusted basis. These results are consistent with
the notion that only some managers have skill and these managers decide how
to apply their skill depending on the economic environment.
We identify characteristics of these superior funds and their managers. The
superior funds tend to be smaller and more active. By matching fund-level to
manager-level data, we find that the skilled managers are more likely to attract
new money flows and are also more likely to depart later in their careers to
hedge funds. Presumably, both of these patterns are market-based reflections
of skilled managers’ ability.
We entertain many alternative explanations for our main findings. First,
we consider whether mechanical effects from cyclical fluctuations in means or
variances of stock returns could generate the observed patterns in picking and
timing measures. After all, expected stock returns vary with the state of the
business cycle (e.g., Ferson and Harvey (1991) and Dangl and Halling (2012)).
Second, we entertain the possibility that fund strategies change because the
fund manager changes. Third, we analyze potential selection effects at both
the fund and the manager levels. Fourth, we consider whether various forms
of career concerns might explain our results. Fifth, we explore whether skill
changes are a volatility or dispersion effect, rather than a business cycle effect.
Finally, we study whether our results reflect a composition effect and find that
this is not the case. The same manager who picks stocks well in booms also
times the market well in recessions. In short, none of these alternatives can
explain the observed changes in fund portfolios over the business cycle.
Next, we analyze several investment strategies that managers use to time
the market. We find that, on average, they hold more cash in recessions, their
portfolios have lower market betas, and they tend to engage in sector rotation
by investing more money in defensive industries in recessions and in cyclical
industries in booms. All three results suggest that managers actively adjust
their investment behavior over the business cycle.
Finally, our findings point to a new metric to identify skilled managers. We
propose a skill index for each mutual fund defined as a weighted average of that
fund’s market-timing and stock-picking metrics. The weight on market timing
is the real-time probability of a recession, while the weight on stock picking
is the complementary probability. This weighting scheme intuitively empha-
sizes the fund’s market-timing prowess as recessions become more likely and
its stock-picking ability when the likelihood of recession fades away. This skill
index can be constructed in real time, on a monthly basis. We show that a one-
standard-deviation increase in the skill index is associated with 2.3% higher
return performance over the next year, net of expenses and after controlling
for exposure to the market, and 1.1% higher performance after additionally
controlling for size, value, and momentum factor exposures. We then sort all
1458 The Journal of FinanceR

funds into quintiles according to their skill index and track each quintile over
time. We find that the difference in skill index between the highest and lowest
quintiles remains large and positive for up to one year. In contrast, similar dif-
ferences for market timing and stock picking mean-revert quickly. In principle,
similar skill indices could be constructed for hedge funds, other professional
investment managers, or even individual investors.
Our approach is related to studies that link fund performance to business
cycle variation (Ferson and Schadt (1996), Christopherson, Ferson, and Glass-
man (1998), and Moskowitz (2000)). Time variation in fund manager skill is
a useful piece of evidence in the quest to understand fund behavior. Kacper-
czyk, Van Nieuwerburgh, and Veldkamp (2011) find that this skill comes from
managers’ ability to choose portfolios that anticipate micro and macro funda-
mentals. Motivated by this additional evidence, they develop a new information
choice theory of fund management that can explain the time-varying skill facts
and is supported by a host of other evidence. Glode (2011) argues that funds
outperform in recessions because their investors’ marginal utility is highest in
such periods. While complementary to our explanation, and a good explana-
tion for why households choose to delegate their portfolios to mutual funds, this
work remains silent on what strategies investment managers pursue to achieve
this differential performance. Similarly, Kosowski (2011) shows that fund per-
formance varies over the business cycle but does not distinguish between the
sources of skill as we do here. Finally, de Souza and Lynch (2012) investigate
cyclical performance by mutual fund style using a Generalized Method of Mo-
ments (GMM) technique. Our focus is on detecting the time-varying strategies
that skilled funds employ that are behind the cyclical outperformance result.
The rest of the paper is organized as follows. Section I describes our data.
Section II tests the hypothesis that fund managers’ stock-picking and market-
timing skill varies over the business cycle, using the universe of actively man-
aged U.S. equity mutual funds. It also delves more deeply into how managers
pick stocks and time the market. Section III considers alternative explana-
tions not based on time-varying use of skill. Section IV proposes a real-time
skill index and uses it to predict fund returns. Section V concludes.

I. Data and Measurement

We begin by describing our data on active mutual funds, their portfolios,
and their returns. We describe our measures of skill and then use the data to
estimate them in booms and recessions.

A. Data
Our sample builds upon several data sets. We begin with the Center for
Research on Security Prices (CRSP) Survivorship Bias Free Mutual Fund
Database. The CRSP database provides comprehensive information about fund
returns and a host of other fund characteristics, such as size (total net as-
sets), age, expense ratio, turnover, and load. Given the nature of our tests and
 Time-Varying Fund Manager Skill 1459

data availability, we focus our analysis on domestic open-end diversified equity

funds, for which the holdings data are most complete and reliable.5 In addition,
we exclude index funds and sector funds. Since the reported objectives do not
always indicate whether a fund portfolio is balanced, we also exclude obser-
vations on funds that allocate less than 80% of their portfolio to stocks in the
current quarter. For mutual funds with different share classes, we aggregate
all the observations pertaining to different share classes into one observation,
since they have the same portfolio composition.6
To address the possibility of incubation bias,7 we exclude observations for
which the year of the observation is prior to the reported fund starting year
as well as observations for which the names of the funds are missing in the
CRSP database. Incubated funds also tend to be smaller, which motivates us
to exclude funds that in the previous month had less than $5 million in assets
under management or fewer than 10 stocks.
Next, we merge the CRSP mutual fund data with the Thomson Reuters
stock holdings database and the CRSP stock price data using the methodology
of Kacperczyk, Sialm, and Zheng (2008). We are able to match about 95% of the
CRSP funds to the Thomson database. These stock holdings data are collected
both from reports filed by mutual funds with the SEC and from voluntary
reports generated by the funds. During most of our sample period, funds are
required by law to disclose their holdings semiannually. Nevertheless, about
49% disclose quarterly.8 To calculate fund returns, we link reported stock hold-
ings to the CRSP stock database. The resulting sample includes 3,477 distinct
funds and 250,219 fund-month observations. The number of funds in each
month varies between 158 in May 1980 and 1,670 in July 2001.
Finally, we map funds to the names of their managers using information from
CRSP, Morningstar, Nelson’s Directory of Investment Managers, Zoominfo, and
Zabasearch. This mapping results in a sample with 4,267 managers. We also
use the CRSP/Compustat stock-level database, which is a source of information
on individual stock returns, market capitalizations, book-to-market ratios, and

5 We base our selection criteria on the objective codes and on the disclosed asset compositions.

We exclude funds with CRSP Database objective codes: International, Municipal Bonds, Bond and
Preferred, and Balanced. We include funds with the following Investment Company Data, Inc.
(ICDI) objectives: AG, GI, LG, or IN. If a fund does not have any of the above ICDI objectives, we
select funds with the following Strategic Insight objectives: AGG, GMC, GRI, GRO, ING, or SCG.
If a fund has neither a Strategic Insight nor ICDI objective, then we go to the Wiesenberger Fund
Type Code and pick funds with the following objectives: G, G-I, AGG, GCI, GRI, GRO, LTG, MCG,
and SCG. If none of these objectives are available and the fund has the CS policy (Common Stocks
are the main securities held by the fund), then the fund will be included.
6 We sum the total net assets under management (TNA) of share classes. For the qualitative

attributes of funds (e.g., name, objectives, and year of origination), we retain the observation of the
oldest fund. For the other attributes of funds (e.g., returns, expenses, loads), we take the weighted
average, where the weights are the lagged TNAs of each share class.
7 Bias can arise when fund families incubate several private funds and then only make public

the track record of the surviving incubated funds, not the terminated funds.
8 For 4.6% of observations with valid CRSP data, the previous six months of holdings data are

not available.
1460 The Journal of FinanceR

momentum. The aggregate stock market return is the value-weighted average

return of all stocks in the CRSP universe.
We measure recessions using the definition of the NBER business cycle dat-
ing committee. The start of the recession is the peak of economic activity and its
end is the trough. Our aggregate sample spans 312 months of data from Jan-
uary 1980 through December 2005, among which 38 (12%) are NBER recession
months. Section II.B considers alternative recession indicators.

B. Defining Measures of Skill

Skilled investors form portfolios that outperform the average investor. We
measure skill along two dimensions: market timing and stock picking. If an
investor times the market, it means that he is more exposed to the market
portfolio in periods when the realized market return is high and holds less
when the realized market return is low. Similarly, stock picking means holding
more of a stock in periods when that firm’s realized stock return is high. To
capture these ideas, we define the following measures.
For fund j at time t, Timingtj measures how a fund’s holdings of each asset,
relative to the market, comove with the systematic component of the stock
return:

j
N
j
Timingtj = (wi,t − wi,t
m m
)(βi,t Rt+1 ), (1)
i=1

where βi measures the covariance of stock i’s return, Ri , with the market
return, Rm, divided by the variance of the market return. The portfolio weight
j
wi,t is the fraction of fund j’s total assets held in risky asset i at the start of
time t. The market weight wi,t m
is the fraction of total market capitalization in
asset i. The product of βi and Rm measures the systematic component of asset
i’s returns. Asset i’s βi,t is computed using a rolling-window regression model
of asset i’s excess returns on market excess returns, using return data between
month t − 11 and month t. The return Rt+1 m
is the realized return between the
start of period t and the start of period t + 1. This means that the systematic
component of the return is unknown at the time of portfolio formation. Before
the market return rises, a fund with a high timing ability (Timing) overweights
assets that have high betas. Likewise, it underweights assets with high betas
in anticipation of a market decline.
j
Similarly, Pickingt measures how a fund’s holdings of each stock, relative to
the market, comove with the idiosyncratic component of the stock return:


j
N
j j
Pickingt = (wi,t − wi,t
m i
)(Rt+1 − βi,t Rt+1
m
). (2)
i=1

A fund with a high picking ability (Picking) overweights assets that have
subsequently high idiosyncratic returns and underweights assets with low id-
iosyncratic returns.
 Time-Varying Fund Manager Skill 1461

In terms of interpretation, Timing and Picking are expressed in units of re-

turn per month. They are hypothetical portfolio returns based on the beginning-
j
of-period portfolio weights, wit − witm.9 We also note that the summation is
over all assets in fund j’s portfolio (N j ).10 Our results are robust to defin-
ing the measures as the sum over all stocks held by any of the funds in our
sample.
Our Picking and Timing measures are variants of the performance measures
in Grinblatt and Titman (1993) and Daniel et al. (1997). In particular, they
distinguish performance based on aggregate market returns from that based
on the idiosyncratic components of returns. They are different from the mea-
sures developed by Ferson and Schadt (1996), Becker et al. (1999), and Ferson
and Khang (2002), which compute covariances conditional on available public
information, as we use unconditional measures instead. Conceptually, these
measures are different. For example, in Ferson and Schadt (1996), skill means
executing a trading strategy that outperforms a hypothetical investor who com-
bines publicly available information. In this paper, skill means using public or
private information in a way that generates higher risk-adjusted returns. We
think of managers as having to spend limited time and effort acquiring and
processing any type of information, whether it is private or public, firm spe-
cific or aggregate (Sims (2003)). This cognitive ability to process information,
which we call skill, is what allows the manager to construct a high-performance
portfolio. We now show that the nature of that skill varies over time.

II. Skill Varies over Time

A. Main Results
We begin by testing the main claim of the paper, that skilled investment
managers deploy their skills differently over the business cycle. Our aim is to
show that, because managers analyze the aggregate payoff shock in recessions,
they choose portfolio holdings that comove more with the aggregate shock. Con-
versely, in expansions, their holdings comove more with stock-specific informa-
tion. To this end, we estimate the following regression model:
j j j
Pickingt = a0 + a1 Recessiont + a2 Xt + t , (3)

j j
Timingtj = b0 + b1 Recessiont + b2 Xt + εt , (4)

9 We thank the referee for pointing this out. For a related measure of hypothetical portfolio

returns, see Jiang, Yao, and Yu (2007).

10 We note that the market weights of all stocks in fund j’s portfolio do not sum to one. Rather,

the market weights are the product of two variables with different means. This implies that Timing
does not have a cross-sectional mean of zero.
1462 The Journal of FinanceR

where Recessiont is an indicator variable equal to one if the economy in month

t is in recession, as defined by the NBER, and zero otherwise. X is a vector of
fund-specific control variables, including age (natural logarithm of age in years
since inception, log( Age)), size (natural logarithm of total net assets under
management in millions of dollars, log (TNA)), expense ratio (in % per year,
Expenses), turnover rate (in % per year, Turnover), the percentage flow of new
j j j
funds (the ratio of TNAtj − TNAt−1 (1 + Rt ) to TNAt−1 , Flow), and load (the
sum of front-end and back-end loads, and additional fees charged to customers
to cover marketing and other expenses, Load). Also included are fund style
characteristics along the size, value, and momentum dimensions.11 To mitigate
the impact of outliers on our estimates, we winsorize Flow and Turnover at the
1% level. Finally, we demean all control variables so that the constant a0 can
be interpreted as the level of skill in expansions, and a1 indicates how much
skill increases in recessions.
Table I examines the cyclical variation in market-timing and stock-picking
ability. Columns (1) and (2) show that the average market-timing ability across
funds increases significantly in recessions. Since Timing is expressed in units
of monthly returns, columns (1) and (2) imply that our market-timing measure
is 14 basis points per month or 1.67% per year higher in recessions than in
expansions (and zero in expansions). Likewise, columns (3) and (4) show that
stock-picking ability deteriorates substantially in recessions: Picking is 14 basis
points per month or 1.75% per year lower in recessions than in expansions (and
zero in expansions). In sum, we observe meaningful differences in average
market-timing and stock-picking skills across market conditions.
We estimate this and most of our subsequent specifications using a pooled
(panel) regression model, calculating standard errors by clustering along the
fund and time dimensions. This approach addresses the concern that the errors,
conditional on independent variables, might be correlated within the fund and
time dimensions. Because our variable of interest, Recession, is constant across
all fund observations in a given time period, addressing cross-fund correlation
is important. At the same time, this approach generates standard errors that
may well be overly conservative. To ensure the robustness of our results, we
also explore three alternative ways of clustering. First, we only cluster at the
fund level and not along the time dimension. We find that all coefficients on the
NBER recession indicator variable are strongly significant, with much larger
t-statistics (between 28 and 35) in absolute value. Second, we cluster by fund
style. For this exercise, we sort funds into 64 style bins, based on a 4 × 4 × 4
grouping of the size, value, and momentum characteristics of the stocks they

11 The size style of a fund is the value-weighted score of its stock holdings’ quintile scores

calculated based on the stocks’ market capitalizations (1 denotes the smallest size quintile; 5
denotes the largest size quintile). The value style is the value-weighted score of its stock holdings’
quintile scores calculated based on the stocks’ book-to-market ratios (1 denotes the smallest book-
to-market quintile; 5 denotes the largest book-to-market quintile). The momentum style is the
value-weighted score of a fund’s stock holdings’ percentile scores calculated based on the stocks’ past
12-month returns (1 denotes the smallest return quintile; 5 denotes the largest return quintile).
These style measures are similar in spirit to those defined in Kacperczyk, Sialm, and Zheng (2005).
 Time-Varying Fund Manager Skill 1463

Table I
Timing and Picking Skills Are Cyclical
The dependent variables are Timing and Picking, defined in equations (1) and (2), where each
stock’s beta is measured over a 12-month rolling window. Recession is an indicator variable equal
to one for every month the economy is in a recession according to the NBER, and zero otherwise.
Log( Age) is the natural logarithm of fund age in years. Log(TNA) is the natural logarithm of fund
total net assets. Expenses is the fund expense ratio. Turnover is the fund turnover ratio. Flow is
the percentage growth in a fund’s new money. Load is the total fund load. The last three control
variables measure the style of a fund along the Size, Value, and Momentum dimensions, based on
the average scores of stocks in the fund’s portfolio in that month sorted into quintiles along each
respective characteristic. All control variables are demeaned. Flow and Turnover are winsorized
at the 1% level. The data are monthly and cover the period 1980 to 2005. Standard errors (in
parentheses) are clustered by fund and time.

Timing Picking

(1) (2) (3) (4)

Recession 0.140 0.139 −0.144 −0.146

(0.070) (0.068) (0.047) (0.047)
Log(Age) 0.006 0.004
(0.006) (0.004)
Log(TNA) 0.000 −0.003
(0.003) (0.003)
Expenses 0.677 −0.636
(1.150) (0.537)
Turnover 0.008 0.012
(0.011) (0.007)
Flow 0.003 0.044
(0.077) (0.078)
Load 0.066 0.142
(0.178) (0.106)
Size −0.009 0.005
(0.009) (0.007)
Value −0.015 0.030
(0.013) (0.010)
Momentum −0.015 0.034
(0.034) (0.034)
Constant 0.007 0.007 −0.010 −0.010
(0.024) (0.024) (0.018) (0.018)
Observations 221,306 221,306 221,306 221,306

hold. This clustering allows for dependence within each of the 64 style bins.
All coefficients on the NBER recession indicator are more significant, with t-
statistics in excess of 100 in absolute value. Third, we cluster standard errors
at the fund family level. In this estimation, the t-statistics are between 23 and
24 in absolute value. The Internet Appendix provides the detailed results.12
All of these results reinforce the statistical significance of our findings.
The effects of Recession on Timing and Picking are also robust to including
indicator variables for high aggregate volatility and high earnings dispersion.

12 The Internet Appendix may be found in the online version of this article.
1464 The Journal of FinanceR

Changes in Picking and Timing are not statistically significantly related to

stock return dispersion or volatility, once the effect of Recession is controlled
for, and do not diminish the effect of Recession. These results are in the Internet
Appendix.

B. Real-Time Recession Indicators

The previous results use the official NBER turning points to split the sample
into boom and recession months. But how does a manager know when to use a
market-timing strategy, given that NBER recessions are not known until sev-
eral months after the fact? She need not know the NBER turning points. Just
as she forecasts future market returns or future abnormal returns of individual
stocks, she can forecast the future state of the macroeconomy. We might think
of her as updating the probability of recession (estimating a two-state regime
switching model) based on all public and private information that she has gath-
ered and processed, and formulating an investment strategy that is a weighted
average of her market-timing and stock-picking strategies, with weights de-
termined by the estimated real-time recession probability. The econometrician
who wants to assess managers’ ability to conduct this forecasting will want to
know when the recession actually took place, not just when real-time public
information would lead one believe there was a recession. Because the NBER
Business Cycle Dating Committee uses information available well after the
boom or recession has ended, it produces a more accurate assessment of the
state of the business cycle. This makes the NBER recession indicator the best
metric for the econometrician to investigate ex post whether the fund pursued
the right trading strategy at the right time, and thus serves as the basis for
our primary results.
Nevertheless, we also investigate whether our results hold for two alterna-
tive recession indicators that are available in a more timely fashion. They have
an advantage over the NBER recession indicator variable in that they are con-
tinuous measures of the strength of the economy, rather than a coarser discrete
measure. The first, RecRT, is the real-time recession probability measure con-
structed by Chauvet and Piger (2008).13 The second indicator, RecCFNAI, is
the Chicago Fed National Activity Index (CFNAI) multiplied by −1; RecCFNAI
is negative when economic activity is above average and is positive when eco-
nomic activity is below average.14 Table II reports results that are similar to
13 Real-time recession probabilities for the United States are obtained from a dynamic-factor
Markov-switching model applied to four monthly coincident variables: nonfarm payroll employ-
ment, the index of industrial production, real personal income excluding transfer payments, and
real manufacturing and trade sales. An analysis of the performance of this model for dating busi-
ness cycles in real time and more details are in Chauvet and Piger (2008). Results are similar for
the real-time recession probability measure from the Survey of Professional Forecasters (see the
Internet Appendix).
14 The CFNAI is a coincident indicator of national economic activity comprising 85 existing

macroeconomic time series. It is constructed to have an average value of zero and a standard
deviation of one. We use the headline three-month moving average. The CFNAI is released in the
third week of the month following the month to which it pertains.
 Time-Varying Fund Manager Skill 1465

Table II
Timing and Picking with Real-Time Recession Indicators
The dependent variables are Timing and Picking. Rec RT is a recession measure based on the
Chauvet and Piger (2008) real-time recession probability. Rec RT is a continuous variable and is
expressed in %; its mean is 7.52 and its standard deviation is 17. RecCFNAI is the Chicago Fed
National Activity Index, multiplied by −1. RecCFNAI is a continuous variable and has mean of 0.08
and standard deviation of 0.54. All other controls are defined in Table I. All independent variables,
including Rec RT and RecC F N AI, are demeaned in the regression. The data are monthly and
cover the period 1980 to 2005. Standard errors (in parentheses) are clustered by fund and time.

Timing Picking

(1) (2) (3) (4)

RecRT 0.004 −0.002

(0.002) (0.001)
RecCFNAI 0.094 −0.059
(0.058) (0.029)
Log(Age) 0.008 0.007 0.004 0.004
(0.007) (0.007) (0.004) (0.004)
Log(TNA) −0.001 −0.001 −0.003 −0.003
(0.003) (0.003) (0.003) (0.003)
Expenses 0.552 0.524 −0.607 −0.358
(1.186) (1.069) (0.543) (0.550)
Turnover 0.010 0.009 0.012 0.012
(0.011) (0.010) (0.007) (0.007)
Flow −0.002 −0.002 0.042 0.051
(0.078) (0.077) (0.078) (0.079)
Load 0.092 0.093 0.136 0.111
(0.188) (0.176) (0.108) (0.108)
Size −0.008 −0.009 0.006 0.005
(0.008) (0.009) (0.007) (0.007)
Value −0.019 −0.017 0.029 0.028
(0.013) (0.012) (0.010) (0.010)
Momentum −0.019 −0.018 0.036 0.032
(0.035) (0.036) (0.034) (0.033)
Constant 0.019 0.019 −0.022 −0.022
(0.024) (0.024) (0.017) (0.017)
Observations 221,292 221,292 221,292 221,292

those using NBER-defined recessions. An increase in real-time recession prob-

ability Rec RT from 0% to 50% increases Timing by 20 basis points per month
or 2.46% per year and decreases Picking by 12 basis points per month or 1.41%
per year. An increase in RecCFNAI from its mean of zero to a value of 0.7—
CFNAI readings below −0.7 are typically considered recessionary—increases
Timing by seven basis points per month or 79 basis points per year and de-
creases Picking by four basis points per month or 49 basis points per year. The
effects are measured precisely and are of similar magnitude as the effects of
NBER recessions shown in Table I.
1466 The Journal of FinanceR

C. Do All Managers Have Time-Varying Skills?

Since markets have to clear, not everyone can outperform the market. Sharpe
(1991) and Fama and French (2010) use this adding-up constraint to argue
that the average actively managed mutual fund cannot outperform passively
managed funds, and thus the average fund cannot be a profitable stock-picker.
Table I bears this out: average Picking is negative. The same is not true for
market timing, however, since individual investors have negative timing ability
because they systematically buy index funds when returns are low (Savov
(2010)). A second part of the Fama and French argument is that the R2 of a
regression of the aggregate mutual fund return on the market return is close
to one. In other words, when we average across active funds, that average
fund is passive. Our claim is not that all funds outperform, or even that the
average fund outperforms. Rather, we claim that a subset of funds have skilled
managers who deliver valuable services to their clients, before fees, at the
expense of all other investors (unskilled fund and nonfund investors).
If there exists a subset of skilled managers and they choose to deploy dif-
ferent skills over the business cycle, then most of the time we should observe
variation in the use of skill among the most skilled managers. We test this
prediction using the quantiles of the cross-sectional distribution of fund skill.
Our hypothesis is that the distribution of picking and timing skills should be
more sensitive to the recession variable in the right tail than at the median.
Note that this is not a forgone conclusion: while the average Timing of the
top group of funds sorted by Timing is by construction higher than that of the
median fund, the effect of Recession on Timing need not be higher. We eval-
uate this hypothesis formally by estimating the models in equations (3) and
(4) using quantile regressions. We consider three different quantiles: 50th per-
centile (median, Q50), 75th percentile (Q75), and 95th percentile (Q95). In this
regression, standard errors are calculated using block bootstraping (with 2,000
repetitions), which takes into account cross-sectional dependence across funds
(Luetkepohl (1993)). Table III presents the results.
Consistent with our hypothesis, we find that the effect of the business cy-
cle on skill is much stronger for extremely successful fund managers (those
in quantile 95) than for the median fund. The effect is statistically and eco-
nomically significant, both for stock picking and market timing. For example,
the effect of Recession on Timing for extremely successful managers is about
four times larger than that for the median manager, at 25.1 versus 5.9 basis
points per month. The 19 basis point cross-sectional difference translates into
2.3% points per year. A similar comparison for Picking shows that the effect
of Recession doubles at the 95th percentile compared to the 50th percentile.
While the economic magnitudes of the recession effect are stronger for higher
quantiles, estimation error also increases. The t-statistic for Timing increases
from 2.5 to 3.1, going from quantile 50 to quantile 95; at the same time, the
respective t-statistics for Picking are 4.0 and 2.6. We conclude that the effect of
market conditions on skill matters more for top-performing managers, which
is consistent with the view that only a subset of fund managers are skilled.
 Time-Varying Fund Manager Skill 1467

Table III
Cyclical Variation in Timing and Picking in the Cross-Section
of Funds
The dependent variables are Timing and Picking, defined in equations (1) and (2). The independent
variables are the same as in Table I. This table shows results from estimating quantile regression
models at the median (columns (1) and (4), Q50), 75th percentile (columns (2) and (5), Q75), and 95th
percentile (columns (3) and (6), Q95) of the cross-sectional distribution (across funds) of Timing
and Picking. Standard errors are computed using block bootstraping, where the block is a cluster
of analysis as in Luetkepohl (1993).

Timing Picking

Q50 Q75 Q95 Q50 Q75 Q95

(1) (2) (3) (4) (5) (6)

Recession 0.059 0.114 0.251 −0.084 −0.091 −0.173

(0.023) (0.041) (0.082) (0.021) (0.022) (0.067)
Log(Age) 0.000 −0.003 −0.020 0.003 −0.005 −0.057
(0.001) (0.004) (0.017) (0.002) (0.003) (0.010)
Log(TNA) 0.000 0.004 −0.004 −0.001 0.001 0.005
(0.001) (0.003) (0.010) (0.001) (0.002) (0.007)
Expenses 0.162 4.015 21.046 −0.588 3.096 18.869
(0.258) (1.036) (3.464) (0.277) (0.597) (1.842)
Turnover 0.001 0.053 0.404 0.001 0.042 0.305
(0.001) (0.012) (0.042) (0.001) (0.006) (0.031)
Flow 0.004 0.036 0.228 0.035 0.099 0.192
(0.011) (0.048) (0.218) (0.021) (0.041) (0.140)
Load −0.013 −0.327 −1.404 0.108 −0.129 −1.213
(0.028) (0.110) (0.475) (0.036) (0.078) (0.249)
Size 0.000 −0.015 −0.071 0.005 −0.026 −0.130
(0.001) (0.005) (0.026) (0.003) (0.005) (0.019)
Value −0.001 −0.031 −0.172 0.015 −0.006 −0.046
(0.002) (0.010) (0.044) (0.004) (0.006) (0.021)
Momentum −0.001 0.037 0.196 0.013 0.071 0.278
(0.005) (0.019) (0.072) (0.009) (0.013) (0.047)
Constant 0.000 0.108 0.765 −0.015 0.126 0.722
(0.004) (0.020) (0.061) (0.005) (0.013) (0.053)
Observations 221,306 221,306 221,306 221,306 221,306 221,306

D. The Same Manager Exhibits Both Skills

One possible explanation for the findings reported thus far is that some man-
agers have timing ability and others have picking ability, but no manager both
picks stocks and times the market well. To examine whether some managers
are good at both tasks, we test the prediction that the same mutual funds
that exhibit stock-picking ability in expansions display market-timing ability
in recessions. We first identify funds with superior stock-picking ability in ex-
pansions: for all expansion months, we select all fund-month observations that
j
are in the highest 25% of the Pickingt distribution (equation (2)). We then form
the indicator variable Top (Top j ∈ {0, 1}), which is equal to one for the 25% of
funds (884 funds) with the highest fraction of observations (months) in that
1468 The Journal of FinanceR

Table IV
The Same Funds Switch Strategies
We divide all fund-month observations into recession and expansion subsamples. Expansion =
1 − Recession. Top is an indicator variable equal to one for all funds whose Picking in expansion
is in the highest 25th percentile of the distribution, and zero otherwise. Control variables, sample
period, and standard errors are described in Table I.

Timing Picking

Expansion Recession Expansion Recession

(1) (2) (3) (4)

Top −0.001 0.037 0.059 −0.054

(0.004) (0.013) (0.005) (0.017)
Log(Age) 0.009 −0.015 −0.001 0.027
(0.002) (0.006) (0.002) (0.007)
Log(TNA) −0.001 0.004 −0.001 −0.024
(0.001) (0.003) (0.001) (0.003)
Expenses 0.571 0.981 −0.985 −3.491
(0.322) (1.085) (0.366) (1.355)
Turnover 0.010 0.009 0.013 −0.005
(0.003) (0.008) (0.004) (0.012)
Flow 0.058 −0.852 0.127 −0.054
(0.024) (0.112) (0.036) (0.092)
Load 0.124 0.156 0.104 0.504
(0.050) (0.162) (0.054) (0.197)
Size −0.009 −0.057 0.011 0.023
(0.002) (0.006) (0.002) (0.007)
Value −0.018 −0.057 0.027 0.107
(0.003) (0.010) (0.003) (0.011)
Momentum −0.007 −0.148 0.031 −0.007
(0.003) (0.010) (0.004) (0.011)
Constant 0.018 0.055 −0.022 −0.159
(0.001) (0.005) (0.002) (0.006)
Observations 204,311 18,354 204,311 18,354

group, relative to the total number of observations for that fund (months in ex-
pansions). Next we estimate the following pooled regression model separately
for expansions and recessions:
j j j
Abilityt = c0 + c1 Toptj + c2 Xt + t , (5)

where Ability denotes either Timing or Picking, and X is a vector of previously

defined control variables. The coefficient of interest is c1 .
In Table IV, column (3), we confirm that Top funds are significantly better
at picking stocks in expansions, after controlling for fund characteristics. This
is true by construction. In expansions, Picking is 5.9 basis points per month
or 70 basis points per year higher for Top funds than for the remaining funds.
The main point Table IV makes is that the same Top funds are on average also
better at market timing in recessions. This result is evident from the positive
coefficient on Top in column (2), which is statistically significant at the 5% level.
 Time-Varying Fund Manager Skill 1469

In recessions Timing is 3.7 basis points per month or 45 basis points per year
higher for Top funds than for all other funds. Finally, Top funds do not exhibit
superior market-timing ability in expansions (column (1)) or superior stock-
picking ability in recessions (column (4)). The fact that this group of funds
is not better at either strategy all the time confirms that Top funds switch
strategies.
The Internet Appendix shows that the fund manager need not know the
NBER recession indicator to execute this switching strategy. The results in
Table IV are robust to using the two real-time recession variables introduced
in Section II.B.
A final note about Table IV is that the Top group has a significantly lower
value for Picking during recessions (column (4)). In principle, poor stock-picking
performance in recessions could offset the benefits from superior market timing
in recessions and stock-picking in expansions. Studying the performance of the
Top funds, which we turn to next, will be informative about whether the Top
funds are indeed better managers.15
The existence of some skilled mutual funds with cyclical investment strate-
gies is a robust result. First, the results survive if we change the cutoff levels
for inclusion in the Top portfolio. Second, we confirm our results using Daniel
et al.’s (1997) definitions of market timing (CT) and stock picking (CS). Third,
we reverse the sort to show that funds in the top 25% of market-timing ability
in recessions have statistically higher stock-picking ability in expansions and
higher unconditional alphas. All these results are in the Internet Appendix.

E. Fund Skill or Fund Manager Skill?

Is skill embodied in the manager or does it come from human capital and
the organizational setup the fund provides for that manager? To answer this
question, we follow a manager over time and across funds. Columns (1) and
(2) of Table V show how Timing and Picking change in recessions when the
unit of observation is the manager. The results without the control variables
are similar to those with the controls, which we present. The table indicates
significantly higher Timing (16 basis points per month or 1.87% per year) and
significantly lower Picking (19 basis points per month or 2.30% per year) in
recessions. The magnitudes of the recession effect are similar at the manager
level to those at the fund level. In columns (3) and (4), we add manager fixed
effects to control for any unobserved manager characteristics that may drive
the results. The results remain essentially unchanged. The results are also

15 An alternative interpretation of the negative sign of Recession in column (4) is that it may be

due in part to measurement error. Because measurement error can make slope and intercept esti-
mates negatively correlated, the estimated good market-timers in recessions may bias downward
stock picking for the Top group. We thank our referee for pointing this possibility out. However,
such an argument cannot explain our main result, that good stock-pickers in expansions are good
market-timers in recessions because that result is based on two different regressions, one for the
subsample of recession months and one for expansion months.
1470 The Journal of FinanceR

Table V
Managers as the Unit of Observation
The dependent variables are Timing and Picking, defined in equations (1) and (2), both tracked
at the manager level. In columns (3) and (4), we include manager fixed effects. Control variables,
sample period, and standard errors are described in Table I.

Timing Picking

(1) (2) (3) (4)

Recession 0.156 −0.192 0.160 −0.187

(0.074) (0.053) (0.074) (0.054)
Log(Age) 0.005 0.002 0.004 −0.003
(0.005) (0.004) (0.009) (0.007)
Log(TNA) 0.001 −0.001 0.004 0.002
(0.003) (0.004) (0.005) (0.008)
Expenses 0.194 −0.586 0.569 −0.491
(1.205) (0.596) (1.206) (0.803)
Turnover 0.008 0.012 0.004 0.011
(0.013) (0.008) (0.010) (0.008)
Flow 0.014 0.181 0.019 0.166
(0.094) (0.177) (0.095) (0.184)
Load 0.170 0.051 0.223 −0.016
(0.209) (0.121) (0.191) (0.185)
Size −0.009 0.007 −0.015 0.005
(0.010) (0.008) (0.019) (0.011)
Value −0.029 0.030 −0.033 0.021
(0.016) (0.010) (0.019) (0.014)
Momentum −0.022 0.036 −0.026 0.037
(0.039) (0.042) (0.048) (0.048)
Constant 0.012 −0.009 0.012 −0.010
(0.026) (0.021) (0.027) (0.022)
Manager N N Y Y
Fixed Effect
Observations 333,582 333,582 333,582 333,582

robust to using real-time recession measures (see the Internet Appendix). We

conclude that our results hold both at the fund and at the manager levels.

F. Funds that Switch Strategies Earn Higher Returns

If skilled funds switch between market timing and stock picking, then these
strategy switchers should outperform unskilled funds both in recessions and
in expansions. Table IV shows that there exist Top funds that have both high
stock-picking skill in booms and high market-timing skill in recessions. Table
VI compares the unconditional performance of these Top funds to that of all
other funds. The dependent variables are CAPM, three-factor, and four-factor
alphas obtained from 12-month rolling-window regression of a fund’s excess
returns, based on reported fund returns before expenses, on a set of common
risk factors. After controlling for various fund characteristics, we find that the
CAPM, three-factor, and four-factor alphas are four to seven basis points per
 Time-Varying Fund Manager Skill 1471

Table VI
Strategy Switchers Outperform
The dependent variables—CAPM alpha, three-factor alpha, and four-factor alpha—are obtained
from a 12-month rolling-window regression of a fund’s excess returns, before expenses, on a set
of common risk factors. Top is an indicator variable equal to one for all funds whose Picking in
expansion is in the highest 25th percentile of the distribution, and zero otherwise. Control variables,
sample period, and standard errors are described in Table I. Expansion equals one every month
the economy is not in recession according to the NBER, and zero otherwise.

CAPM alpha Three-Factor Alpha Four-Factor Alpha

(1) (2) (3)

Top 0.068 0.040 0.058

(0.028) (0.018) (0.016)
Log(Age) −0.035 −0.029 −0.038
(0.008) (0.006) (0.006)
Log(TNA) 0.036 0.013 0.014
(0.005) (0.004) (0.003)
Expenses 4.972 0.187 0.070
(0.942) (0.777) (0.716)
Turnover −0.002 −0.049 −0.042
(0.012) (0.011) (0.008)
Flow 2.543 1.765 1.608
(0.181) (0.101) (0.101)
Load −0.649 −0.067 −0.282
(0.186) (0.138) (0.146)
Size −0.057 0.001 −0.000
(0.025) (0.008) (0.009)
Value 0.125 −0.061 −0.020
(0.044) (0.025) (0.017)
Momentum 0.296 0.184 0.177
(0.038) (0.028) (0.023)
Constant 0.058 0.041 0.050
(0.020) (0.016) (0.019)
Observations 226,769 226,769 226,769

month or 48 to 82 basis points per year higher for the Top portfolio, a difference
that is statistically and economically significant. These results are the same
order of magnitude as the difference in Timing and Picking between the Top
funds and all other funds, which is consistent with the interpretation of Timing
and Picking as (hypothetical) returns. The return measures in Table VI pro-
vide additional evidence because they are based on observed fund returns, not
hypothetical returns. Finally, given that the Top funds are no better at market
timing in expansions and strictly worse at stock picking during recessions (re-
call Table IV), these unconditional outperformance results show that the Top
funds are following market-timing strategies in recessions and stock-picking
strategies in expansions.
The Internet Appendix shows that the unconditional outperformance of the
Top funds also holds when Top fund membership is defined based on real-time
recession measures. The outperformance is between 5.0 and 7.2 basis points
1472 The Journal of FinanceR

per month, very similar to the baseline results. This evidence supports the
case for a robust link between various recession measures, including real-time
measures, and fund outperformance. De Souza and Lynch (2012) find that the
average fund outperformance in recessions is not robust to ex ante measures
of recession. Our results show that the unconditional outperformance of the
group of highly skilled funds is present regardless of an ex ante or ex post
definition of recession.

G. The Characteristics of Skilled Funds and Managers

In Panel A of Table VII, we compare characteristics of the funds in the Top
portfolio to those of funds not included in this portfolio. We note several differ-
ences. First, funds in the Top portfolio are younger (by five years on average).
Second, they have less wealth under management (by $400 million), sugges-
tive of decreasing returns to scale at the fund level. Third, they tend to charge
higher expenses (by 0.26% per year), suggesting rent extraction from customers
for the skill they provide. Fourth, they exhibit higher portfolio turnover (130%
per year, vs. 80% for other funds), consistent with a more active management
style. Fifth, they receive higher inflows of new assets to manage, presumably
a market-based reflection of their skill. Sixth, they tend to hold portfolios with
fewer stocks and higher stock- and industry-level portfolio dispersion, mea-
sured as the Herfindahl index of portfolio weights in deviation from the market
portfolio’s weights. Seventh, their betas deviate more from their peers, sug-
gesting a strategy with different systematic risk exposure. Finally, they rely
significantly more on aggregate information. Taken together, fund characteris-
tics such as age, TNA, expenses, and turnover explain 14% of the variation in
the skill indicator Top (see the Internet Appendix for these results). Including
attributes that we could link to skilled funds’ active investment behavior, such
as stock and industry portfolio dispersion and beta deviation, increases the R2
to 19%.
Table VII, Panel B, examines manager characteristics. Managers of Top funds
are 2.6% more likely to have an MBA, are one year younger, and have 1.7 fewer
years of experience. Interestingly, they are much more likely to depart for hedge
funds later in their careers, suggesting that the market judges them to have
superior skill. Taken together, these findings paint a rough picture of what a
typical skilled fund looks like.

H. Market Timing: Varying Cash or Betas?

Next, we explore in greater detail how managers time the market. A fund
manager can time the market even if she only holds the market portfolio of risky
assets. For example, if the manager invests 100% of her assets in the S&P 500
when market returns are high and holds only cash when the market is falling,
j
she will score high on timing ability because her weight wit will be high in
booms and zero in market downturns. She can also time the market without
holding any cash by holding a high-beta portfolio (of stocks or industries) in
booms and a low-beta portfolio in downturns. We find that managers do some
 Time-Varying Fund Manager Skill 1473

Table VII
Comparing Top Funds to Other Funds
We divide all fund-month observations into recession and expansion subsamples. Expansion equals
one every month the economy is not in recession according to the NBER, and zero otherwise. Top is
one for any fund with Picking (defined in Table I) in the highest 25th percentile in expansions, and
zero otherwise. Panel A reports fund-level characteristics. Age, TNA, Expenses, Turnover, and Flow
are defined in Table I. RSI comes from Kacperczyk, Sialm, and Zheng (2008). Portfolio Dispersion
is the concentration of the fund’s portfolio, measured as the Herfindahl index of portfolio weights in
deviation from the market portfolio’s weights. Stock Number is the number of stocks in the fund’s
portfolio. Industry is the industry concentration of the fund’s portfolio, measured as the Herfindahl
index of portfolio weights in a given industry in deviation from the market portfolio’s weights. Beta
Deviation is the absolute difference between the fund’s beta and the average beta in its style
category. Panel B reports manager-level characteristics. MBA or Ivy equals one if the manager
obtained an MBA degree or graduated from an Ivy League institution, and zero otherwise. Age and
Experience are the fund manager’s age and experience in years. Gender equals one if the manager
is a male and zero if female. Hedge Fund equals one if the manager departed to a hedge fund, and
zero otherwise. Top1–Top0 is the difference between the mean values of the groups for which Top
equals one and zero, respectively. p-values measure statistical significance of the difference. The
data are monthly from 1980 to 2005.

Top = 1 Top = 0 Difference

Mean Stdev. Median Mean Stdev. Median Top1–Top0 p-value

Panel A: Fund Characteristics

Age 10.01 8.91 7 15.20 15.34 9 −5.19 0.000

TNA 621.13 2,027.04 129.60 1,019.45 4,024.29 162.90 −398.32 0.002
Expenses 1.48 0.47 1.42 1.22 0.47 1.17 0.26 0.000
Turnover 130.41 166.44 101.00 79.89 116.02 58.00 50.52 0.000
Flow 0.22 7.39 −0.76 −0.07 6.47 −0.73 0.300 0.008
Portfolio Dispersion 1.68 1.60 1.29 1.33 1.50 0.99 0.35 0.000
Stock Number 90.83 110.20 68 111.86 187.13 69 −21.03 0.000
Industry 8.49 7.90 6.39 5.37 7.54 3.54 3.12 0.000
Beta Deviation 0.18 0.38 0.13 0.13 0.23 0.10 0.05 0.000
RSI 4.13 5.93 1.82 2.77 3.97 1.26 1.37 0.000

Panel B: Fund Manager Characteristics

MBA 42.09 49.37 0 39.49 48.88 0 2.60 0.128

Ivy 25.36 43.51 0 27.94 44.87 0 −2.57 0.205
Age 53.02 10.42 50 54.11 10.06 52 −1.08 0.081
Experience 26.45 10.01 24 28.14 10.00 26 −1.69 0.003
Gender 90.89 28.77 100 90.50 29.31 100 0.39 0.681
Hedge Fund 10.43 30.57 0 6.12 23.96 0 4.31 0.000

of each: in recessions, they significantly increase their cash holdings, reduce

their holdings of high-beta stocks, and tilt their portfolios away from cyclical
and towards more defensive sectors.
To investigate changes in cash holdings, we measure cash either as Reported
Cash from CRSP or Implied Cash, backed out from fund size and its equity
holdings. In expansions, funds hold about 5.25% of their portfolios in cash.
In recessions, the fraction of their cash holdings rises by about 3 percentage
points for Implied Cash and by 0.4 percentage points for Reported Cash. Both
1474 The Journal of FinanceR

increases are statistically significant with t-statistics around five, and each
represents a change of about 10% of a standard deviation of the dependent
variable. We also investigate the month-over-month change in the Implied
Cash position. In recessions, cash holdings increase by 0.5 percentage points
while in expansions they fall by 1.5 percentage points. The effect of Recession
is modest, but measured precisely. Within one year of the end of the average
recession, half of the Implied Cash buildup is reversed (1.5% of the 3%).
Second, we examine whether fund managers invest in lower beta stocks in
recessions. For each individual stock, we compute the beta (from 12-month
rolling-window regressions). Based on the individual stock holdings of each
mutual fund, we construct the funds’ (value-weighted) equity beta. This beta is
1.11 in expansions and 1.00 in recessions; the 0.11 difference has a t-statistic of
four. This means that funds hold different types of stocks in recessions, namely,
lower beta stocks.16
Finally, we investigate whether funds rotate their portfolio allocations across
different industries over the business cycle. In recessions, funds increase their
portfolio weights (relative to those in the market portfolio) in low-beta sectors
such as Healthcare, Nondurables (which includes Food and Tobacco), Whole-
sale, and Utilities. They reduce their portfolio weights (relative to those in the
market portfolio) in high-beta sectors such as Telecom, Business Equipment
and Services, Manufacturing, Energy, and Durables.17 Hence, funds engage in
sector rotation over the course of the business cycle in a way consistent with
market timing.
In sum, funds time the market by lowering their portfolio beta, shifting
to defensive sectors, and increasing their cash positions in recessions. The
Internet Appendix reports the complete set of results for each of these exercises.

III. Alternative Explanations

This section explores whether our time-varying skill results could arise from
composition effects, or from other effects unrelated to managerial skill.

A. Composition Effects
Suppose that each fund pursues a fixed strategy, but the composition of
funds changes over the business cycle in such a way as to make the average
fund strategy change. Such composition effects could come from changes in the
set of active funds, from changes in the size of each of those funds, or from
entry and exit of fund managers. We explore each of these possibilities in turn
and show that they do not drive our results.

16 The results on cash holdings and equity betas are robust to using real-time recession measures

(see the Internet Appendix).

17 While the sector weights are persistent over time, we can reject the null hypothesis of a unit

root in the regression residuals for each of the sectors using the test developed by Maddala and
Wu (1999). Hence, persistent regressor bias is unlikely to explain these results.
 Time-Varying Fund Manager Skill 1475

A.1. Fund-Level Composition Effects

First, we rerun our results with fund-fixed effects to control for changes in the
set of active funds. Including fixed effects in a regression model is a standard
response to sample selection concerns. The results are qualitatively similar
and slightly stronger quantitatively. For example, the coefficient on Recession
in the Picking equation is equal to −0.146 (identical to the estimate without
fixed effects), while the recession coefficient in the Timing estimation is slightly
higher 0.148 (as opposed to 0.139 before). Both coefficients are significant at
the 1% level of statistical significance

A.2. Size-Driven Composition Effects

Next, we consider whether composition related to fund size could drive our
effect. Mutual funds might change their strategies over the business cycle only
because relative fund size changes. In particular, some fund managers might
become more successful in recessions and manage larger funds, while others
become successful in booms and accumulate more assets in those times. Our
results showing that the same funds that do well at stock picking in expansions
are good at market timing in recessions (Table IV) are incompatible with this
explanation. Furthermore, if this effect holds, it should also be picked up with
fund fixed effects. Yet, when we include fund fixed effects, our cyclical skill
results persist.

A.3. Manager-Level Composition Effects

Similarly, we can rule out the alternative explanation that the composition
of managers changes over the cycle; recall our manager-level results with man-
ager fixed effects as explanatory variables (columns (3) and (4) of Table V). If
a selection/composition effect drives the increase in Timing in recessions, we
should not find any effect from recession once we control for fixed effects. How-
ever, our results show that all our manager-level results survive the inclusion
of manager fixed effects.
More specifically, if we think that the composition of managers is changing
over the business cycle through entry and exit of managers, we should see
some difference in observable manager characteristics.18 However, when we
examine manager characteristics over the business cycle, we find no systematic
differences in age, experience, or educational background of fund managers in
recessions versus expansions.

18 Our data show that investment fund managers’ outside labor market options deteriorate

in recessions. Not only do assets under management—and therefore managerial compensation—

shrink, but managers are also more likely to get fired or demoted. There is a smaller incidence of
promotion to a larger mutual fund in a different fund family, a higher incidence of demotion to a
smaller mutual fund in a different fund family, and a lower incidence of departure to a hedge fund.
See the Internet Appendix for these results.
1476 The Journal of FinanceR

B. Stock Price Patterns Generate Mechanical Effects

Our results at the mutual fund level could arise mechanically from the prop-
erties of returns at the stock level. To rule this possibility out, we generate ar-
tificial return data for a panel of 1,000 stocks and the same number of periods
as our sample. We assume that stock returns follow a CAPM with time-varying
parameters. The mean and volatility of the market return, the idiosyncratic
volatility, and the cross-sectional standard deviation of the alpha and beta are
chosen to match the properties of stock-level data. Using a simulation for 500
funds, we verify that mechanical mutual fund strategies cannot reproduce the
observed features of fund returns. The mechanical strategies include: (1) an
equally weighted portfolio of 75 (or 50 or 100) randomly chosen stocks by all
funds; (2) half the funds choosing 75 random stocks from the top half of the
alpha distribution and the other half choosing 75 stocks from the bottom half
of the alpha distribution; (3) similar strategies in which half the funds pick
from the top half of the total return or the beta distribution, with the other half
of funds choosing from the bottom half of the distribution. While this exercise
does not consider every single alternative mechanical strategy, none of these
strategies generates higher market-timing measures in recessions and higher
stock-picking readings in expansions.

C. Career Concerns
We next consider the possibility that fund behavior changes over the business
cycle because of cyclical career concerns. Chevalier and Ellison (1999) show
that career concerns give managers an incentive to herd. This pressure is
strongest for young managers. It would seem logical that concerns related
to being fired would be greatest in recessions; our data bear this out (see
footnote 18). What does herding imply for picking and timing? Stock picking
is an activity that skilled managers might do very differently: some might
analyze pharmaceutical stocks and others energy stocks. But market timing
is something that managers would expect other skilled managers to do in the
same way at the same time and thus is better suited to herding. So, according
to this alternative explanation, market timing in recessions arises because of
stronger pressure on young managers to herd.
To investigate this hypothesis, we estimate portfolio dispersion—an inverse
measure of herding—in recessions and booms. Our measure of dispersion is
the sum of the squared deviations of fund j’s portfolio weight in asset i at time
j
t, wit , from the average fund’s portfolio weight in asset i at time t, witm, summed
over all assets held by fund j, N j :


j
N
j  j 2
Port f olio Dispersiont = wit − witm . (6)
i=1
 Time-Varying Fund Manager Skill 1477

If we regress this dispersion measure on a recession indicator variable and

a constant, the recession coefficient is 0.347, significant at the 5% level.19 Con-
trolling for the fund characteristics listed in Table I changes this estimate by
less than 1%. Thus, instead of finding more herding in recessions, we find less.20
While labor market considerations may be important to understand many
aspects of mutual fund managers’ behavior, they do not account for the patterns
we document.

IV. Identifying Skilled Managers in Real Time

The second contribution of the paper is to use the results on time-variation
in skill presented thus far to construct an indicator identifying the skilled
managers. We exploit the prediction that skilled managers time the market in
recessions and pick stocks in expansions to develop Skill Index. Unlike in the
previous sections, we now take the perspective of an investor (or an agency like
Morningstar) who wants to form a timely gauge of funds’ skill. Our monthly
Skill Index is constructed based on real time, publicly available information.
We show that this index is correlated with future performance. Second, we show
that, unlike market-timing or stock-picking alone, the Skill Index is persistent
over time.

A. Creating Skill Index

To identify skilled investment managers, it is important that these managers
can be identified in real time, without the benefit of looking at the full sample
of the data. To this end, we construct Skill Index, which is informed by our
main result that the nature of skill and investment strategies change over the
business cycle.
We define Skill Index for fund j in month t + 1 as a weighted average of
j
Timingtj and Pickingt , in which the weights we place on each measure depend
on the state of the business cycle:
j
Skill Indext+1 = wt Timingtj + (1 − wt )Pickingtj . (7)

19 This portfolio dispersion measure is similar in spirit to the concentration measure used in

Kacperczyk, Sialm, and Zheng (2005) and the active share measure used in Cremers and Petajisto
(2009).
20 It is worth noting that we do find that manager age is positively significantly related to the

fund’s portfolio dispersion, meaning that younger managers are more likely to herd. This confirms
the findings of Chevalier and Ellison (1999) in our data set. But this herding is weaker in recessions,
not stronger. Since we just showed that recessions are times when managers are more likely to
deviate from the pack, one might be tempted to construct a story whereby career concerns are
actually stronger in expansions instead of recessions. But if that is true, then younger managers
should hold portfolios with lower dispersion in booms. When we regress portfolio dispersion on
recession, age of the manager, and the interaction of recession with age, the interaction term
should have a negative sign (dispersion for older managers decreases less in recessions). Instead,
we find a significantly positive interaction effect of 0.40 with a standard error of 0.08.
1478 The Journal of FinanceR

We normalize Timing and Picking so that each has a mean of zero and
a standard deviation of one in the cross-section, each period. Next, we set
the weight on Timing equal to 0 ≤ wt ≤ 1, where wt is the real-time recession
probability of Chauvet and Piger (2008). This continuous weighting scheme
is quite intuitive: linearly weight Timing more whenever the probability of a
recession increases. Picking always gets the complementary weight 1 − pt . The
resulting Skill Index is mean zero with standard deviation close to one (0.96).
Notice that Timingtj and Pickingtj are both constructed using fund portfolio
weights at the beginning of time t and asset returns realized between the start
of period t and t + 1. All of this information is known at time t + 1. Also, the
real-time recession probability pt is known at time t + 1. Thus, this is a fund
score that can be computed at the end of each period t + 1 and contains no future
information (beyond time t + 1) that would generate spurious predictability.

B. Returns by Skill Index

In the first exercise, we sort each fund into one of five quintiles based on
the Skill Index each month, ranked from low skill to high skill. For each quin-
tile portfolio, we compute equal-weighted average portfolio returns. Since we
now take the perspective of the investor, alphas are measured based on re-
ported returns net of expenses.21 This creates a portfolio return time series for
each quintile of the Skill Index distribution. We then estimate a time-series
regression of quintile excess returns on the aggregate market excess return
(CAPM), size and value factor returns (three-factor model), and momentum
factor returns (four-factor model). The four panels of Table VIII show the aver-
age abnormal fund return, CAPM, three-factor, and four-factor alphas over the
1-, 3-, 6-, 9-, and 12-month periods post–portfolio formation (not cumulative
returns).
Performance measures increase monotonically with Skill Index (from Q1 to
Q5). While the average fund does not outperform (last row of each panel), the
average fund in the top quintile of Skill Index performs substantially better
than the average fund in the bottom quintile. The annual return difference is
3% to 6%, depending on the measure of performance, one to six months after
portfolio formation, and 0.6% to 2.1% 12 months after portfolio formation. In
sum, the strategy delivers economically significant spread returns net of fees.
The advantage of these portfolio sorting results is that they are based on time-
series regressions, which make no use of overlapping data. The disadvantage is
that they do not control for fund characteristics. In a second exercise, we control
for fund characteristics but use overlapping data. In particular, we examine
whether Skill Index at time t + 1 can predict fund performance, measured by
the CAPM, three-factor, and four-factor fund alphas one month ahead (based
on returns realized between time t + 1 and time t + 2) or one year ahead (based

t+1 depends on wt , Timingt , and Pickingt . The

21 Consistent with equation (7), the Skill Index

portfolio return for each quintile is formed as the equal-weighted average of fund returns between
t + 1 and t + 2 for all funds in that quintile of the Skill Index distribution.
 Time-Varying Fund Manager Skill 1479

Table VIII
Skill Index Portfolio Sorts
Each month we sort mutual funds into five quintiles based on their Skill Index, defined in equation
(7), from lowest values (Q1) to highest values (Q5). We report equal-weighted average abnormal
returns (Panel A), CAPM alphas (Panel B), Fama-French three-factor alphas (Panel C), and Fama-
French-Carhart four-factor alphas (Panel D) of each quintile portfolio 1, 3, 6, 9, or 12 months after
portfolio formation. The last row of each table (average) reports the average abnormal return or
alpha across all funds. All numbers represent monthly returns (in %).

Panel A: Abnormal Returns Panel C: FF 3F Alpha

1 mo 3 mo 6 mo 9 mo 12 mo 1 mo 3 mo 6 mo 9 mo 12 mo

Q1 −0.262 −0.076 −0.240 −0.049 −0.017 Q1 −0.176 −0.180 −0.193 −0.198 −0.088
Q2 −0.121 −0.059 −0.199 −0.094 −0.041 Q2 −0.101 −0.101 −0.104 −0.110 −0.070
Q3 −0.052 −0.047 −0.065 −0.094 −0.064 Q3 −0.067 −0.065 −0.062 −0.065 −0.065
Q4 0.033 −0.015 0.074 −0.006 −0.032 Q4 −0.019 −0.018 −0.020 −0.018 −0.059
Q5 0.250 0.037 0.263 0.081 −0.007 Q5 0.117 0.108 0.101 0.095 −0.025
Q5−Q1 0.512 0.113 0.502 0.130 0.009 Q5−Q1 0.293 0.288 0.294 0.293 0.064
Average −0.030 −0.032 −0.033 −0.032 −0.032 Average −0.049 −0.051 −0.055 −0.059 −0.061

Panel B: CAPM Alpha Panel D: Carhart 4F Alpha

1 mo 3 mo 6 mo 9 mo 12 mo 1 mo 3 mo 6 mo 9 mo 12 mo

Q1 −0.271 −0.289 −0.283 −0.304 −0.107 Q1 −0.144 −0.146 −0.152 −0.164 −0.067
Q2 −0.138 −0.146 −0.146 −0.161 −0.086 Q2 −0.083 −0.085 −0.086 −0.086 −0.053
Q3 −0.052 −0.050 −0.054 −0.061 −0.061 Q3 −0.055 −0.054 −0.051 −0.054 −0.053
Q4 0.064 0.071 0.063 0.068 −0.014 Q4 −0.010 −0.009 −0.014 −0.015 −0.047
Q5 0.249 0.253 0.242 0.262 0.065 Q5 0.113 0.106 0.096 0.091 −0.016
Q5−Q1 0.519 0.541 0.526 0.566 0.172 Q5−Q1 0.257 0.252 0.248 0.255 0.051
Average −0.029 −0.032 −0.035 −0.039 −0.041 Average −0.036 −0.037 −0.041 −0.045 −0.047

on returns realized between time t + 1 and time t + 13).22 Table IX shows that
funds with a higher Skill Index have higher average net alphas. For example,
when Skill Index is at its mean of zero, the net alpha is around −0.5% per year.
However, when Skill Index is one standard deviation (0.96) above its mean,

22 One-month-ahead alphas are obtained from time-series rolling-window regressions of fund

returns on standard style benchmarks. For example, we estimate a CAPM regression from 12-
j j
month rolling-window regressions of fund returns on the market return: Rt+2 = α j + β j Rt+2m
+ t+2 .
The 12-month estimation window runs from t − 10 until t + 2, where time t + 1 denotes the time at
which Skill Indext+1 is constructed and known. We then define the one-month-ahead alpha as the
j j
part of the return not explained by covariation with the market: αt+2 = α̂ j + t+2 = Rt+2 − β̂ j Rt+2
m
.
This is the analogue of an abnormal fund return except that it takes into account the possibility
that the fund’s beta with the market may not be unity. The inclusion of the idiosyncratic return
j
piece t+2 is standard in the literature. While the constant α j is estimated with return information
j
that is partially known at time t + 1, t+2 is not measurable with respect to time t + 1 information.
In practice, most of the variation in the one-month-ahead alpha in the panel regression arises from
j
the t+2 term. The one-year-ahead alphas use return information from t + 1 to t + 13 to estimate α̂ j
j
and add t+13 . The one-year-ahead results generate very similar point estimates to the one-month
results, something that would be highly unlikely if the one-month-ahead alphas were severely
biased due to look-ahead issues or mechanical correlations.
1480 The Journal of FinanceR

Table IX
Skill Index Predicts Performance
The dependent variables are, respectively, the fund’s cumulative CAPM, three-factor, or four-factor
alpha, calculated from a 12-month rolling-window regression. The regression window is t − 10 to
t + 2 for one month ahead and t + 1 to t + 13 for one year ahead. For each fund, we form Skill Index
defined in equation (7). Picking and Timing are defined in Table I, except that they are normalized
so that they are mean zero and have a standard deviation of one over the full sample. The other
control variables, sample period, and standard error calculation are the same as in Table I.

One Month Ahead One Year Ahead

CAPM Three-Factor Four-Factor Alpha CAPM Three-Factor Four-Factor

Alpha Alpha Alpha Alpha Alpha Alpha
(1) (2) (3) (4) (5) (6)

Skill Index 0.202 0.103 0.094 0.197 0.090 0.091

(0.038) (0.019) (0.017) (0.028) (0.023) (0.013)
Log(Age) −0.027 −0.022 −0.033 −0.014 −0.008 −0.023
(0.007) (0.005) (0.006) (0.007) (0.005) (0.006)
Log(TNA) 0.025 0.005 0.008 −0.012 −0.018 −0.012
(0.004) (0.003) (0.003) (0.004) (0.003) (0.003)
Expenses −3.347 −8.139 −8.040 −5.571 −9.423 −9.475
(1.026) (0.797) (0.755) (0.983) (0.748) (0.660)
Turnover −0.041 −0.075 −0.065 −0.007 −0.050 −0.048
(0.010) (0.010) (0.008) (0.012) (0.011) (0.009)
Flow 2.226 1.585 1.436 0.106 0.163 0.164
(0.156) (0.095) (0.091) (0.114) (0.084) (0.071)
Load −0.655 −0.037 −0.271 −0.576 0.250 −0.009
(0.189) (0.134) (0.143) (0.174) (0.122) (0.132)
Size −0.031 0.016 0.012 −0.061 −0.005 −0.005
(0.024) (0.008) (0.009) (0.028) (0.010) (0.010)
Value 0.237 0.010 0.045 0.235 0.034 0.074
(0.030) (0.019) (0.017) (0.036) (0.025) (0.021)
Momentum 0.246 0.158 0.157 0.098 0.056 0.088
(0.042) (0.031) (0.025) (0.031) (0.030) (0.025)
Constant −0.032 −0.056 −0.042 −0.044 −0.071 −0.058
(0.023) (0.017) (0.020) (0.024) (0.018) (0.021)
Observations 219,321 219,321 219,321 187,659 187,659 187,659

the one-month-ahead CAPM alpha is 2.4% higher per year. The three- and
four-factor alphas are, respectively, 1.2% and 1.1% points higher per year for a
one-standard-deviation increase in Skill Index. The three right-most columns
show similar predictive power of Skill Index for one-year-ahead alphas. A one-
standard-deviation increase in Skill Index is associated with 2.2% per year
higher CAPM alpha and 1.0% higher three- and four-factor alphas.
The Internet Appendix shows that these results are robust to using different
definitions for Skill Index. In particular, they explore a different weighting
scheme for wt in equation (7). Both measures set wt equal to 0.8 in recessions
and 0.2 in expansions. But the first measure defines recessions as months
with real-time recession probabilities above 20% while the second measure
 Time-Varying Fund Manager Skill 1481

defines recessions as months in which CFNAI is below −0.7. Both results are
qualitatively and quantitatively similar to the benchmark results.

C. Persistence of Skill Measures

While skill is persistent, luck is not. The fact that stock picking and mar-
ket timing do not exhibit much persistence casts doubt on the existence of
fund manager skill.23 To show this, we first sort funds into quintiles based on
their Timing scores in month 0 and track their performance over the next 1 to
12 months. We then subtract the average Timing of funds that were initially
in quintile 1 (Q1) from that of funds that were initially in quintile 5 (Q5). We
do the same for funds sorted by their stock-picking scores. The top two panels
of Figure 1 plot the Q5–Q1 differences in skill scores over these 12 months. If
skill is persistent, we should see the top market-timers (stock-pickers) in month
0 continue to outperform the worst month 0 market-timers (stock-pickers) in
months 1, 2, and beyond. Instead, the difference in market-timing (top panel)
and stock-picking (middle panel) skill disappears, even just one month postfor-
mation. On average, the previous month’s worst market-timers are no worse
than the previous month’s best market-timers.
However, our Skill Index captures more general cognitive ability that is
more flexible, that is, one that can be applied to picking stocks successfully one
month and to timing the market in other months, or to doing some of each. If
there exist skilled managers but they employ different skills at different times,
that could explain why neither picking nor timing is persistent. But this more
general measure of ability should be. To test this conjecture, we perform the
same sorting exercise on Skill Index. The bottom panel of Figure 1 reveals that
managers with high Skill Index in one month on average still display higher
skill 12 months later. This difference is statistically significant for up to six
months.

V. Conclusion
Do investment managers add value for their clients? The answer to this
question has implications for discussions ranging from market efficiency to
practical portfolio advice for households. The large amount of randomness in
financial asset returns and the unobservable nature of risk make this a difficult
question to answer. Most previous studies ignored the fact that the type of skill
funds exhibit might change with the state of the business cycle. When we
condition on the state of the business cycle, we find that managers successfully
pick stocks in booms and time the market in recessions. Managers who exhibit
this time-varying skill outperform the market by 50 to 90 basis points per year.

23 Their first-order autocorrelation coefficients are not statistically different from zero. This lack

of persistence also alleviates the concern that the results in Table I suffer from spurious regression
bias. Formal tests of the null hypothesis that the errors from panel regressions (3) and (4) contain
a unit root, due to Maddala and Wu (1999), are rejected at the 1% level.
1482 The Journal of FinanceR

Figure 1. Persistence of Timing, Picking, and Skill Index. We rank funds into quintiles
based on their Timing, Picking, or Skill Index score at time 0. Next, we subtract the average score
in quintile 5 (Q5) from that in quintile 1 (Q1) in each of the following 12 months. We report that
difference in the postformation period. A positive difference indicates persistent skill. The shading
shows two standard errors on either side of the point estimate (solid line).

Our findings raise the question: why do skilled fund managers change the na-
ture of their activities over the business cycle? Kacperczyk, Van Nieuwerburgh,
and Veldkamp (2011) provide a theoretical answer to this question. They argue
that recessions are times when aggregate payoff shocks are more volatile and
when the price of risk is higher. Both of these forces make acquiring and pro-
cessing information about aggregate shocks more valuable. Thus, if a firm has
general cognitive ability that it can allocate between processing information
about specific stocks or processing information about the aggregate economy, it
will optimally change the allocation across booms and recessions. Thus, our ap-
proach uncovers new evidence in support of the idea that a subset of managers
process information about firm-specific and economy-wide shocks in a way that
creates value.
Our findings leave several interesting questions for future research. One such
question is why the group of funds that we associate with superior performance
does not raise fees or attract inflows until outperformance disappears. We do
observe higher fees, smaller size, and higher inflows for this group of funds,
suggesting that to some extent the equalizing forces operate. Their strength is
 Time-Varying Fund Manager Skill 1483

likely mitigated by the presence of trading costs (including the inability to short
poorly performing mutual funds), partial investor unawareness of the patterns
we document, and uncertainty about the economic environment. Given the
volatility of stock and fund returns, it takes time for investors to identify the
best funds and for fund managers to learn about their own ability. Future work
could fruitfully incorporate such considerations.

Initial submission: November 16, 2011; Final version received: May 28, 2013
Editor: Campbell Harvey

REFERENCES
Amihud, Yakov, and Ruslan Goyenko, 2013, Mutual fund’s R2 as predictor of performance, Review
of Financial Studies 26, 667–695.
Avramov, Doron, and Russ Wermers, 2006, Investing in mutual funds when returns are predictable,
Journal of Financial Economics 81, 339–377.
Baker, Malcolm, Lubomir Litov, Jessica Wachter, and Jeffrey Wurgler, 2010, Can mutual fund
managers pick stocks? Evidence from their trades prior to earnings announcements, Journal
of Financial and Quantitative Analysis 45, 1111–1131.
Becker, Connie, Wayne Ferson, David Myers, and Michael Schill, 1999, Conditional market timing
with benchmark investors, Journal of Financial Economics 52, 119–148.
Bollen, Nicolas P., and Jeffrey A. Busse, 2001, On the timing ability of mutual fund managers,
Journal of Finance 56, 1075–1094.
Chauvet, Marcelle, and Jeremy Piger, 2008, A comparison of the real-time performance of business
cycle dating methods, Journal of Business and Economic Statistics 26, 42–49.
Chevalier, Judith, and Glenn Ellison, 1999, Career concerns of mutual fund managers, Quarterly
Journal of Economics 104, 389–432.
Christoffersen, Susan E.K., Donald B. Keim, and David K. Musto, 2007, Valuable information and
costly liquidity: Evidence from individual mutual fund trades, Working paper, University of
Toronto and University of Pennsylvania.
Christopherson, Jon, Wayne Ferson, and Debra Glassman, 1998, Conditioning manager alphas on
economic information: Another look at the persistence of performance, Review of Financial
Studies 11, 111–142.
Cohen, Randolph, Christopher Polk, and Bernhard Silli, 2011, Best ideas, Working paper, London
School of Economics.
Cremers, Martijn, and Antti Petajisto, 2009, How active is your fund manager? A new measure
that predicts performance, Review of Financial Studies 22, 3329–3365.
Dangl, Thomas, and Michael Halling, 2012, Predictive regressions with time-varying coefficients,
Journal of Financial Economics 106, 157–181.
Daniel, Kent, Mark Grinblatt, Sheridan Titman, and Russ Wermers, 1997, Measuring mutual fund
performance with characteristic based benchmarks, Journal of Finance 52, 1035–1058.
de Souza, Andre, and Anthony Lynch, 2012, Does mutual fund performance vary over the business
cycle? Working paper, New York University.
Elton, Edwin J., Martin J. Gruber, and Christopher R. Blake, 2012, An examination of mutual
fund timing ability using monthly holdings data, Review of Finance 16, 619–645.
Fama, Eugene F., and Kenneth R. French, 2010, Luck versus skill in the cross section of mutual
fund returns, Journal of Finance 65, 1915–1947.
Ferson, Wayne E., and Campbell R. Harvey, 1991, The variation of economic risk premiums,
Journal of Political Economy 99, 385–415.
Ferson, Wayne E., and Kenneth Khang, 2002, Conditional performance measurement using port-
folio weights: Evidence for pension funds, Journal of Financial Economics 65, 249–282.
Ferson, Wayne E., and Meijun Qian, 2004, Conditional performance evaluation, revisited, in Re-
search Foundation Monograph (CFA Institute, New York).
1484 The Journal of FinanceR

Ferson, Wayne E., and Rudi W. Schadt, 1996, Measuring fund strategy and performance in chang-
ing economic conditions, Journal of Finance 51, 425–461.
French, Kenneth, 2008, The cost of active investing, Journal of Finance 63, 1537–1573.
Glode, Vincent, 2011, Why mutual funds underperform? Journal of Financial Economics 99, 546–
559.
Graham, John R., and Campbell R. Harvey, 1996, Market timing ability and volatility implied in
investment newsletters’ asset allocation recommendations, Journal of Financial Economics
42, 397–421.
Grinblatt, Mark, and Sheridan Titman, 1993, Performance measurement without benchmarks: An
examination of mutual fund returns, Journal of Business 66, 47–68.
Huang, Jennifer, Clemens Sialm, and Hanjiang Zhang, 2011, Risk shifting and mutual fund per-
formance, Review of Financial Studies 24, 2575–2616.
Jiang, George J., Tong Yao, and Tong Yu, 2007, Do mutual funds time the market? Evidence from
portfolio holdings, Journal of Financial Economics 86, 724–758.
Kacperczyk, Marcin, and Amit Seru, 2007, Fund manager use of public information: New evidence
on managerial skills, Journal of Finance 62, 485–528.
Kacperczyk, Marcin, Clemens Sialm, and Lu Zheng, 2005, On the industry concentration of actively
managed equity mutual funds, Journal of Finance 60, 1983–2012.
Kacperczyk, Marcin, Clemens Sialm, and Lu Zheng, 2008, Unobserved actions of mutual funds,
Review of Financial Studies 21, 2379–2416.
Kacperczyk, Marcin, Stijn Van Nieuwerburgh, and Laura Veldkamp, 2011, Rational attention
allocation over the business cycle, Working paper, New York University.
Koijen, Ralph, S. J., 2014, The cross-section of managerial ability, incentives, and risk preferences,
Journal of Finance 69, 1051–1098.
Kosowski, Robert, 2011, Do mutual funds perform when it matters most to investors? U.S. mutual
fund performance and risk in recessions and expansions, Quarterly Journal of Finance 1,
607–664.
Luetkepohl, Helmut, 1993, Introduction to Multiple Time Series Analysis (Springer-Verlag, Berlin).
Maddala, Gangadharrao S., and Shaowen Wu, 1999, A comparative study of unit root tests with
panel data and a new simple test, Oxford Bulletin of Economics and Statistics 61, 631–652.
Mamaysky, Harry, Matthew Spiegel, and Hong Zhang, 2008, Estimating the dynamics of mutual
fund alphas and betas, Review of Financial Studies 21, 233–264.
Moskowitz, Tobias J., 2000, Discussion of mutual fund performance: An empirical decomposition
into stock-picking talent, style, transactions costs, and expenses, Journal of Finance 55, 1695–
1704.
Pástor, Luboš, and Robert F. Stambaugh, 2002, Investing in equity mutual funds, Journal of
Financial Economics 63, 315–349.
Savov, Alexi, 2010, Free for a fee: The hidden cost of index fund investing, Working paper, New
York University.
Sharpe, William F., 1991, The arithmetic of active management, Financial Analyst Journal 47,
7–9.
Sims, Christopher, 2003, Implications of rational inattention, Journal of Monetary Economics 50,
665–690.

Supporting Information
Additional Supporting Information may be found in the online version of this
article at the publisher’s website:
Appendix S1: Internet Appendix.