Anton JF 2014

THE JOURNAL OF FINANCE • VOL. LXIX, NO.
3 • JUNE 2014
Connected Stocks
MIGUEL ANTÓN and CHRISTOPHER POLK∗
ABSTRACT
We connect stocks through their common active mutual fund owners. We show that the
degree of shared ownership forecasts cross-sectional variation in return correlation,
controlling for exposure to systematic return factors, style and sector similarity, and
many other pair characteristics. We argue that shared ownership causes this excess
comovement based on evidence from a natural experiment—the 2003 mutual fund
trading scandal. These results motivate a novel cross-stock-reversal trading strategy
exploiting information contained in ownership connections. We show that long-short
hedge fund index returns covary negatively with this strategy, suggesting these funds
may exacerbate this excess comovement.
Barberis and Shleifer (2003) and Barberis, Shleifer, and Wurgler (2005) ar-
gue that institutional features may play an important role in the movement of
stocks’ discount rates, causing returns to comove above and beyond the comove-
ment implied by their fundamentals. In this paper, we propose a new approach
to document that type of institutional-based comovement. Based on mounting
evidence since Coval and Stafford (2007) that mutual fund flows result in price
pressure, we focus on connecting stocks through active mutual fund owner-
ship. Specifically, we forecast cross-sectional variation in return correlation for
stock pairs using the degree of common ownership by active mutual funds.
Our bottom-up approach allows us to measure institutional-driven comove-
∗ Miguel Antón is with Department of Finance, IESE Business School. Christopher Polk is with
Department of Finance, London School of Economics. We are grateful to Ken French, David Hsieh,
and Bob Shiller for providing us with some of the data used in this study. We are deeply in-
debted to two anonymous referees, an Associate Editor, and the Editor, Cam Harvey, for numerous
comments that significantly improved the paper. We are also grateful for comments from Andriy
Bodnaruk, John Campbell, Randy Cohen, Jonathan Cohn, Owen Lamont, Augustin Landier, Dong
Lou, Narayan Naik, Belén Nieto, Jeremy Stein, Dimitri Vayanos, Tuomo Vuolteenaho, and Paul
Woolley, as well as from conference participants at the Summer 2008 LSE lunchtime workshop,
Spring 2009 Harvard PhD brownbag lunch, 2010 HEC 2nd Annual Hedge Fund Conference, 2010
Paul Woolley Research Initiative Workshop, 2010 Spanish Finance Forum, 2010 WFA, 2010 EFA,
2010 Yale University Whitebox Graduate Student Conference, and 2010 Paul Woolley Conference,
and seminar participants at Bristol University, IESE-ESADE, Imperial College, London School of
Economics, and the University of Amsterdam. Financial support from the Paul Woolley Centre at
the LSE is gratefully acknowledged. Antón also gratefully acknowledges support from the Fun-
dación Ramón Areces, the European Commission (Project EMAIFAP–FP7-PEOPLE-2011-Marie
Curie CIG (GA no. 303990)), and the Public-Private Sector Research Center at IESE Business
School.
DOI: 10.1111/jofi.12149
1099
15406261, 2014, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/jofi.12149 by University Of Macau Wu Yee Sun, Wiley Online Library on [30/07/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1100 The Journal of FinanceR
ment more precisely. For example, we can isolate abnormally connected firms
by controlling for a host of pair characteristics.
We find that cross-sectional variation in common ownership predicts higher
four-factor abnormal return correlation, controlling for similarity in industry,
size, book-to-market, and momentum characteristics as well as the degree of
common analyst coverage. Our analysis also explores cross-sectional varia-
tion in the strength of the effect. In particular, we document that common
ownership has a stronger effect on subsequent correlation when the common
owners are experiencing strong net flows either into or out of their funds. The
flow effect is particularly strong when the stocks in the pair have relatively
low float.
The effect we document is economically as well as statistically significant
even though we restrict our analysis to big stocks, that is, stocks with market
capitalization above the NYSE median value. In fact, when we take into account
cross-sectional variation in the strength of the effect, we find that the predicted
variation in the four-factor residual correlation on average ranges from –0.046
to 0.029.
The fact that fund ownership is endogenous is a key concern. Perhaps fund
owners merely invest in stocks that have common fundamentals and thus nat-
urally comove. Indeed, many, if not all, papers arguing that fund ownership
causes comovement are vulnerable to this criticism.1 We are the first to ad-
dress this critique based on evidence from a natural experiment, the 2003
mutual trading scandal.2 This unexpected event resulted in significant out-
flows to implicated funds, which we show results in exogenous variation in
common ownership. We document that this exogenous variation strongly fore-
casts variation in abnormal return correlation. Thus, we provide novel evidence
of a causal relation between stock connectedness and comovement.
Previous and current research looks at related questions: Is there informa-
tion in institutional holdings about future returns? Or, more specifically, does
variation in assets under management result in price pressure? Most of these
studies are concerned with cross-sectional and time-series predictability of ab-
normal returns. Any implications for comovement are secondary, if examined
at all. We begin by measuring comovement. We then turn to the implications
for predictability of returns at the end of the analysis. In particular, we use the
1 For example, Vijh (1994) and Barberis, Shleifer, and Wurgler (2005) find that S&P 500 index
additions are followed by an increase in return covariation. However, members of the S&P 500
selection committee might implicitly look to include firms in the index that, going forward, are cen-
tral to the economy. Intuitively, that implicit criterion might be plausibly related to the sensitivity
of a firm’s fundamentals to the fundamentals of other central firms. Indeed, Antón (2011) provides
evidence that this is the case. Similarly, Greenwood and Thesmar (2011) show that their measure
of co-fragility (based on funds’ correlated flows) forecasts subsequent comovement. However, funds
with correlated flows might choose to invest in similar stocks whose fundamentals comove. Indeed,
Greenwood and Thesmar (2011) grant that their work has serious endogeneity concerns, and thus
their results are only suggestive.
2 Kisin (2011) uses this scandal to explore determinants of firm investment.
Connected Stocks 1101
return on stocks connected to a particular firm to predict cross-sectional varia-

tion in average returns. Specifically, we define a stock’s connected return as the
return on a portfolio of all the stocks in our sample abnormally connected to
the stock through common ownership at that point in time. By using the actual
connected return as a measure of the price impact from mutual fund trading,
we avoid having to measure the impact of flows on stock returns.3
We use the return on a stock’s connected portfolio as a confirming sig-
nal that the stock has been temporarily pushed away from true value
by mutual fund trading. Our cross-stock-reversal trading strategy gener-
ates significant abnormal returns of more than 9% per year, controlling
for market, size, value, momentum, and the own-stock, short-term reversal
factors.4
Finally, we use our connected-return strategy to explain hedge fund index re-
turns in standard performance attribution regressions. We show that the long-
short hedge fund index loads negatively on our trading strategy, particularly
when the VIX index is increasing. In fact, the exposure of the value-weighted
long-short hedge fund index is more negative than the corresponding expo-
sure of a value-weighted portfolio of the active mutual funds in our sample.
This finding suggests that the typical hedge fund may be part of the problem
(causing the comovement) instead of part of the solution.
In summary, we show that understanding connectedness is a simple way
to identify institutional-based stock comovement and its link to cross-stock-
reversal patterns. The rest of the paper is organized as follows. In Section I,
we summarize the literature. In Section II, we describe our methodology and
data sources. Section III presents our results. Section IV concludes.
I. Literature Review
Our work has an intellectual link to papers studying whether crisis periods
result in sudden increases in correlations across countries, a feature often
referred to as contagion.5 In particular, we follow Bekaert, Harvey, and Ng
3 Researchers must make several assumptions to back out flows from data on funds’ total net
assets (see, e.g., Chevalier and Ellison (1997), Sirri and Tufano (1998), and Lou (2012)). These
assumptions include that inflows and outflows occur at the end of each period (often as infrequently
as quarterly), and that investors reinvest their dividends and capital appreciation distributions in
the same fund. Lack of data on flow activity during fund initiation, merger, and liquidation is also
an issue. Furthermore, the relation between flows and stock-level price impact is complicated by
the fact that funds can absorb capital flows using cash buffers and can initiate, expand, or liquidate
individual positions to different degrees based on liquidity costs.
4 As our focus is on the economics of how connectedness causes excess comovement that, by
definition, must eventually revert, we do not analyse whether our trading strategy is costly to
implement. Though the strategy is restricted to big stocks and the predictability persists for six
months, we grant that the relatively high turnover of the strategy will make the performance after
transaction costs less attractive.
5 See, for example, King and Wadhwani (1990), King, Sentana, and Wadhwani (1994), Forbes
and Rigobon (2001), Rigobon (2002), and Bekaert, Harvey, and Ng (2005).
(2005, p. 40) and define contagion as “excess correlation, that is, correlation
over and above what one would expect from economic fundamentals.” Like
Bekaert, Harvey, and Ng, we take an asset pricing perspective to measure
economic fundamentals and identify contagion by the correlation of an asset
pricing model’s residuals. While Bekaert, Harvey, and Ng examine national
equity market returns, we look instead at firm-level stock returns.
Our work is directly related to a growing literature on fund flows. That
mutual fund flows and past performance are related (Ippolito (1992), Chevalier
and Ellison (1997), Sirri and Tufano (1998)) is now well known. A recent paper
by Coval and Stafford (2007) documents that extreme flows result in forced
trading that temporarily moves prices away from fundamental value as in
the general asset fire-sales model of Shleifer and Vishny (1992) through the
price pressure mechanism of Scholes (1972). Ellul, Jotikasthira, and Lundblad
(2010) and Mitchell, Pedersen, and Pulvino (2007) document broadly similar
findings in the bond and convertible bond markets, respectively.6 Unlike these
papers, which study particular events, our analysis explores the extent to which
institutional connections affect second moments more generally.
Recent theoretical work emphasizes the importance of delegated portfolio
management and agency frictions for price movements such as these.7 In par-
ticular, Vayanos and Woolley (2013) show that fund flows can generate comove-
ment and lead-lag effects of the type we document. Their model provides strong
theoretical motivation for our empirical analysis. More generally, beginning
with Shleifer and Vishny (1997), researchers have studied the role of funding
in arbitrage activity and the extent to which arbitrageurs should be expected to
demand or provide liquidity.8 In a related study, Sadka (2010) shows that the
typical hedge fund loads on a liquidity risk factor and that sensitivity to that
liquidity risk is priced in the cross section of hedge fund returns. Measuring
the extent to which hedge funds’ performance can be attributed to a trading
strategy that exploits temporary price dislocations resulting from institutional-
driven comovement follows naturally from this theory and empirical evidence.
Two contemporaneous papers analyze issues related to comovement and
institutional ownership.9 First, Lou (2012) argues that flow-driven demand
6 Similarly, Singleton (2011) examines the influence of investor flows on returns in the crude oil
futures market.
7 See, for example, the 2010 American Finance Association presidential address, Duffie (2010).
8 Many researchers have built on the ideas in Shleifer and Vishny (1997), including Gromb and
Vayanos (2002), Vayanos (2004), and Brunnermeier and Pedersen (2009). For a recent survey of
this literature, see Gromb and Vayanos (2010).
9 Sun (2007) uses standard clustering techniques to identify subsets of funds that hold similar
stocks. Sun shows that the typical stock’s return covaries with the equal-weighted average return
on all of the stocks in the top five fund clusters holding the stock in question. Moreover, Sun shows
that this covariance is stronger if the average flow for the top five clusters in question is lower
than the tenth percentile of the historical distribution of fund flows for that group of five fund
clusters. In contrast, our approach predicts the pair-specific return correlation with the degree
of the pair’s common ownership, controlling for exposure to systematic return factors, style and
sector similarity, and many other pair characteristics. Additionally, Sun does not examine any
implications of the covariance she documents for profitable trading strategies.
shocks more generally affect prices than just in the extreme fire-sale situa-
tions of Coval and Stafford (2007). Lou shows that stocks with high expected
flows from mutual funds comove. Of course, one might expect that mutual fund
flows are to similar stocks that might naturally covary, so endogeneity is an
issue.
Lou (2012) also examines the predictability of returns linked to mutual fund
flows. Unlike Lou (or Coval and Stafford (2007) for that matter), we avoid
having to measure the impact of flows on stock returns and instead use the
actual connected return as a signal of the strength of the contagion effect
resulting from ownership-based connections in the stock market.10 Moreover,
whereas Lou’s focus is on momentum effects, we instead examine the way the
presence of institutional connectedness generates cross-stock-reversal patterns
in returns.
Second, Greenwood and Thesmar (2011) argue that active mutual fund own-
ers of stocks can have correlated trading needs and thus the stocks that they
hold can comove, even if there are no overlapping holdings. Greenwood and
Thesmar show that these correlated trading needs predict future price volatil-
ity and cross-sectional variation in comovement. Greenwood and Thesmar
acknowledge that a concern with their interpretation is that funds with corre-
lated flows might simply choose to invest in similar stocks whose fundamentals
comove.
Researchers have adopted our techniques to study comovement in other mar-
kets. For example, Jotikasthira, Lundblad, and Ramadorai (2012) use our
method to argue that investor flows into global funds affect equity prices,
emerging market return correlations, and the correlation between emerging
markets and developed markets. Bartram, Griffin, and Ng (2012) study con-
nected stocks using international holdings data. They develop a similar mea-
sure of ownership linkages and show that return shocks linked to ownership
connections are as important in explaining firm-level stock returns as industry
and country returns.
Finally, like us, Chen et al. (2008) explore whether hedge funds take ad-
vantage of the mutual fund flow-forced trading that Coval and Stafford (2007)
document. They argue that hedge funds take advantage of that opportunity
as average returns of long-short hedge funds are higher in months when the
number of mutual funds in distress is large. In particular, Chen et al. suggest
that this evidence is consistent with hedge funds front-running the trades of
distressed mutual funds. Our findings are consistent with their results but
further show that the typical hedge fund apparently winds up on the wrong
side of the price dislocation that we study.
10 In a similar fashion, Lou and Polk (2013) propose using the past degree of abnormal return
correlation among those stocks that an arbitrageur would speculate on as a measure of arbitrage
activity. Their approach is successful in linking time-variation in both the profitability and subse-
quent reversal of momentum strategy returns to their measure of momentum arbitrage activity.
II. Data and Methodology

A. Data and Sample
Stock returns, trading volume, and other relevant market and accounting
data come from the intersection of the CRSP daily and monthly files and COM-
PUSTAT. We restrict our analysis to common stocks (share codes 10 and 11)
from NYSE, Amex, and NASDAQ whose market capitalizations are above the
NYSE median market cap (i.e., “big” stocks).11 We choose these screening cri-
teria because common ownership by active managers is not pervasive—small
stocks, especially in the beginning of the sample, have little institutional own-
ership. Limiting the data in this way also keeps the sample relatively homo-
geneous and ensures that the patterns we find are not just due to small or
microcap stocks (Fama and French (2008)).
Data on mutual fund holdings come from merging the CDA/Spectrum
database provided by Thomson Reuters and the CRSP Mutual Fund database.
We use the Mutual Fund Links data set created by Russ Wermers and provided
by Wharton Research Data Services. As our focus is on U.S. active mutual
funds, we remove index funds, tax-managed funds, and international funds by
applying standard screening criteria used in the literature.12 We obtain data
on analyst forecasts from the Institutional Brokers Estimate System (I/B/E/S)
database. Our sample covers the period 1980 to 2008.
B. Measuring Common Ownership

At each quarter-end, we measure common ownership as the total value of
stock held by the F common funds of the two stocks, scaled by the total mar-
ket capitalization of the two stocks, and label this variable FC APi j,t . Thus,
F f f
(S Pi,t +S Pj,t ) f
FC APi j,t = f =1 i,t j,t
Si,t Pi,t +S j,t Pj,t
, where Si,t is the number of shares held by fund f
at time t trading at price Pi,t with total shares outstanding of Si,t , and similarly
for stock j. We define common funds as those funds that held both stocks i and j
in their portfolios at the end of quarter t. For each cross section, we calculate the
normalized (to have zero mean and unit standard deviation) rank-transformed
FC APi j,t , which we denote as FC APi∗j,t .
We primarily identify our findings from cross-sectional variation in common
ownership. Indeed, by rank-transforming FCAP, we hope to create a variable
that is comparable across cross sections. On the one hand, this choice means
that our analysis is not obviously going to be dominated by the later cross
sections in which there are many more mutual funds. On the other hand, time-
series variation in the number of mutual funds may help us identify the link
between common ownership and excess comovement. Therefore, we also ex-
plore time-series variation in our effect by examining trends and subperiod
variation in the series of Fama and MacBeth (1973) coefficients. Additionally,
11 Choosing this particular cutoff is broadly consistent with the empirical literature. See, for
example, Fama and French (2008).

12 We specifically follow the algorithm described in Cremers and Petajisto (2009).
we show that our finding of a relation between common ownership and comove-
ment is robust to not rank-transforming FCAP.
C. Modeling Cross-Sectional Variation in Comovement

We estimate cross-sectional regressions forecasting the within-month real-
ized correlation (ρi j,t+1 ) of each stock pair’s daily four-factor abnormal returns
with FC APi∗j,t and a host of pair characteristics that we use as controls:

n
ρi j,t+1 = a + b f ∗ FC APi∗j,t + bk ∗ CONTROLi j,k + εi j,t+1 . (1)
k=1
Rather than pool the data, we estimate these regressions monthly and report
the time-series average as in Fama and MacBeth (1973) as a robust way to
avoid any issues with cross-correlation in the residuals of equation (1). We
then calculate Newey and West (1987) standard errors of the Fama–MacBeth
estimates that take into account autocorrelation in the time series of cross-
sectional estimates out to four lags.13
D. Controls
Our ultimate goal is to show that common ownership causes stocks to become
more correlated. However, being able to forecast differences in comovement
with common ownership may not be surprising if the predictability simply re-
flects the fact that fund managers choose to hold stocks that are similar. Those
stocks would be expected to comove regardless of who owns the stock. The
prototypical example is industry classification; we expect firms in similar in-
dustries to covary more, all else equal. Certainly fund managers may also tend
to invest in a particular sector. To capture that tendency, we measure industry
similarity as the number of consecutive SIC digits (NUMSIC), beginning with
the first digit, that are equal for a given pair.
Similarly, managers may follow investment styles and/or strategies such as
growth or value, small cap or large cap, momentum or reversal. Since previous
research by Fama and French (1993) and Carhart (1997) documents the link
between these characteristics and sensitivity to common return factors, we
expect higher correlation between two stocks if they have a greater similarity
in the aforementioned characteristics. By forecasting the correlation of Fama–
French–Carhart residuals, we hope that a good portion of the variation has
already been removed.14
13 This choice is consistent with the degree of autocorrelation generally observed in the time
series of coefficients. In particular, the partial serial correlation in the time series of regression
coefficients associated with FCAP becomes statistically insignificant after four months. Moreover,
we reestimated our standard errors using lag lengths as long as 36 months, and our results remain
statistically significant.
14 We obtain the Fama and French (1993) and Carhart (1997) daily return factors from Ken
French’s website.
However, since style characteristics do not line up perfectly with regression

loadings (Daniel and Titman (1997)), we also control for similarity in charac-
teristics directly. To measure this similarity each quarter, we first calculate
each stock’s percentile ranking on a particular firm characteristic. Our mea-
sures of similarity, SAMESIZE, SAMEBEME, and SAMEMOM, are then just
the negative of the absolute difference in percentile ranking for a particular
characteristic across a pair.
Presumably, analysts specialize in covering similar stocks. As a consequence,
we create measures of common analyst coverage. Specifically, we measure the
number of analysts Ai j,t that issued at least one annual earnings forecast for
stocks i and j during the 12-month period preceding t.15
Chen, Chen, and Li (2012) document that variables other than similarity
in the aforementioned style characteristics forecast cross-sectional variation
in pair-wise correlations. Their variables include various measures of past
correlation: the past five-year monthly return correlation, RETCORRi j,t ; the
past profitability correlation, ROECORRi j,t ; and the past correlation in the two
stocks’ abnormal trading volume, VOLCORRi j,t . Their variables also include
differences in key firm characteristics: the absolute value of the difference in
five-year log sales growth rate, DIFFGROWTHi j,t ; the absolute difference in
financial leverage ratio (defined as long-term debt/total assets), DIFFLEV i j,t ;
and the absolute value of the difference in the two stocks’ log share price,
DIFFPRICEi j,t . Finally, Chen, Chen, and Li include dummies capturing im-
portant similarity between the two stocks in a pair. These controls include
a dummy variable for whether the two firms are located in the same state,
DSTATEi j,t ; a dummy variable for whether the two stocks belong to the S&P
500 index, DINDEX i j,t ; and a dummy variable for whether the two stocks are
on the same stock exchange, DLISTINGi j,t . We include all of these variables as
controls in many of our specifications.
As with our common ownership measure, we update these controls at each
quarter-end, and, except for the dummy variables of Chen, Chen, and Li (2012),
we do not use these variables directly but instead work with normalized rank
transforms, which we continue to denote with an asterisk superscript. Finally,
as institutional ownership is correlated with size, we also create very general
size controls based on the normalized rank-transform of the percentile mar-
ket capitalization of the two stocks, SIZE1 and SIZE2 (where we label the
larger stock in the pair as the first stock), and the interaction between these
normalized size rankings.
III. Results
A. Forecasting Comovement
Table I, Panel A confirms the well-known marked increase in active funds
over this period. Our choice of restricting the analysis to big stocks results in a
15 We base our measure of common coverage on annual forecasts, as quarterly earnings forecasts
are not issued as consistently.

Table I
Summary Statistics
This table reports summary statistics for all NYSE-Amex-NASDAQ stocks that are above the
median NYSE market capitalization as of each quarter-end and the active funds that invest in
those stocks from 1980 to 2008. Panel A lists the total number of stocks, pairs of stocks, and funds
at the end of the fourth quarter for the first and last
years of the sample, and for every five years.
The number of unique stock pairs is n ∗ (n − 1) /2, where n is the number of stocks. We have
41,374,135 pair-quarters in our sample. Panel B reports summary statistics on the number of
stocks held by each fund and the number of funds that hold each stock, for both the full sample
and each decade within the sample. Panel C reports the distribution of common fund ownership
(FC APi j,t ), which measures the total value of stocks held by all common funds of the two stocks,
scaled by the total market capitalization of the two stocks, as of the quarter-end. The distribution
is shown for the average of the entire sample (All), for the first and the last year of the sample,
and for every five years.
Panel A: Number of Stocks, Pairs, and Funds
Year Stocks Pairs Funds
1980 895 400,065 171

1985 828 342,378 251
1990 809 326,836 469
1995 905 409,060 1,002
2000 902 406,351 1,780
2005 803 322,003 1,901
2008 746 277,885 1,642
Panel B: Summary Statistics for Stocks and Funds
Variable Period Mean Median SD Min Max
Num stocks per fund 1980–2008 58.5 41 69.8 1 1028

1980–1989 40.4 33 35.1 1 439
1990–1999 54.4 40 61.7 1 932
2000–2008 64.0 44 77.7 1 1028
Num funds per stock 1980–2008 68.6 39 84.1 1 829

1980–1989 14.1 10 14.2 1 167
1990–1999 58.0 42 56.3 1 603
2000–2008 141.8 114 103.1 1 829
Panel C: The Cross-Sectional Distribution of Common Fund Ownership
Percentiles
Year Mean SD 0% 25% 50% 75% 95% 99% 100%
All 0.0077 0.0142 0.0000 0.0000 0.0020 0.0086 0.0356 0.0691 0.8143
1980 0.0016 0.0045 0.0000 0.0000 0.0000 0.0005 0.0099 0.0224 0.0970
1985 0.0018 0.0045 0.0000 0.0000 0.0000 0.0015 0.0103 0.0218 0.1013
1990 0.0035 0.0072 0.0000 0.0000 0.0005 0.0038 0.0168 0.0335 0.1447
1995 0.0080 0.0141 0.0000 0.0004 0.0022 0.0091 0.0355 0.0679 0.2755
2000 0.0119 0.0176 0.0000 0.0012 0.0051 0.0148 0.0485 0.0828 0.2581
2005 0.0152 0.0185 0.0000 0.0031 0.0080 0.0199 0.0545 0.0859 0.2164
2008 0.0156 0.0182 0.0000 0.0037 0.0088 0.0205 0.0531 0.0856 0.2922
relatively stable number of stocks over time (approximately 800 to 900 stocks)
and a large cross section of stock pairs to analyze (roughly 300,000 to 400,000
pairs). Table I, Panel B documents that the typical active manager holds 58.5
big stocks. As a consequence, a big stock has on average 68.6 active mutual
fund owners. Of course, because of the growth of funds over this period, these
full-sample numbers mask a strong trend in the number of funds holding a
stock. In the early part of the sample (1980 to 1989), the median number of
funds holding a typical big stock is 10. In the later part of the sample (2000 to
2008), the median increases to 114.
Our specific interest is how these numbers translate to the number of com-
mon owners for a pair of stocks. We report snapshots of the distribution of
common owners in Table I, Panel C. The sharing of active fund ownership with
another stock is relatively common as more than 75% of all stock pairs have a
common active fund owner. A typical pair in our sample has roughly nine funds
in common, which results in 0.77% of the combined market capitalization being
owned by common funds. One might expect from the increase in funds over this
period that the number of ownership-based connections among big stocks has
also increased dramatically. We find this to be the case. In 1990, the median
number of ownership connections is one, owning 0.05% of the combined mar-
ket equity. In 2008, the median number of ownership connections is 19, owning
0.88% of the combined market equity. Our use of rank-transformed variables
in the main analysis ensures that these sorts of trends do not drive our results.
Table II, Panel A reports results from forecasting cross-sectional variation
in four-factor residual correlation for the full sample.16 In the first column,
specification (1), we estimate a simplified version of equation (1) with only
common ownership, FCAP*, as a forecasting variable. We find that FCAP* is
highly statistically significant, with a coefficient of 0.00395 and a t-statistic of
13.43.17 The effect is economically significant as well, with fitted values ranging
from an average minimum of –0.0007 to an average maximum 0.0124 around
an average abnormal correlation of 0.0053.18
In the second column, specification (2), we report results of regressions incor-
porating our controls for style/sector similarity and common coverage. Recall
that these control variables are normalized to have a standard deviation of one
and are transformed so that higher values indicate greater style similarity. The
number of common analysts, A*, forecasts abnormal return correlation with a
statistically significant coefficient of 0.01437 (t-statistic of 11.92). This finding
16 One may worry that we miss some portion of excess comovement between returns due to
non-synchronicity. Our results become slightly stronger if we include lags when computing the
correlation of the daily four-factor residuals.
17 Of course, a portion of the covariation linked to a pair’s exposure to systematic risk factors
could be caused by common ownership. If we instead forecast the correlation of raw returns (i.e.,
not four-factor residuals), the coefficient is 0.02118, more than five times as large.
18 To calculate the average range in abnormal correlation linked to FCAP*, we first orthogonalize
FCAP* to the other variables. We then forecast four-factor residual correlations using orthogonal-
ized FCAP*, saving the minimum and maximum forecast in each cross section. Finally, we average
this range over time.
Table II
Connected Comovement
This table reports Fama and MacBeth (1973) estimates of monthly cross-sectional regressions
forecasting the correlation of daily Fama and French (1993)–Carhart (1997) residuals in month
t + 1 for the sample of stocks defined in Table I. The independent variables are updated quarterly
and include our measure of institutional connectedness, the total ownership value held by all
common funds of the two stocks scaled by the total market capitalization of the two stocks, FC APi j,t ,
and a series of controls at time t. We measure the negative of the absolute value of the difference
in size, book-to-market ratio (BE/ME), and momentum percentile ranking across the two stocks in
the pair (S AMESI ZEi j,t , S AMEBMi j,t , and S AMEMOMi j,t , respectively). We also measure the
number of similar SIC digits beginning with the first digit, NUMSICi j,t , for the two stocks in a
pair. We also include 39 other controls, which are reported in the Internet Appendix, for the sake
of brevity. All independent variables, excluding dummy variables are then rank-transformed and
normalized to have unit standard deviation, which we denote with *. We calculate Newey and West
(1987) standard errors (four lags) of the Fama and MacBeth (1973) estimates that take into account
autocorrelation in the cross-sectional slopes. We report the associated t-statistics in parentheses.
We report estimates of regressions using various subsets of these variables in Panel A. Panel B
shows the same set of regressions as in Panel A, but for three different subsamples (corresponding
to the three decades of the sample). In Panel C, we report the results of regressing the time series
of Fama and MacBeth (1973) FCAP* coefficients on a constant and a trend. In Panels B and C, only
the FCAP* coefficient and the intercept estimates are shown; the remaining controls are reported
in the Internet Appendix.
Panel A: Full Sample (1980–2008)
Dependent Variable: Correlation of 4F Residuals
(1) (2) (3) (4)
Constant 0.00508 0.00512 0.00284 0.00288

(11.30) (11.17) (6.92) (6.85)
FCAP* 0.00395 0.00256 0.00168 0.00184
(13.43) (11.61) (8.58) (9.85)
A* 0.01437 0.01342 0.01334
(11.92) (11.83) (11.77)
SAMESIZE* −0.00365 −0.00396 −0.00402
(−1.43) (−1.53) (−1.54)
SAMEBM* 0.00031 −0.00024 −0.00001
(2.68) (−2.80) (−0.00)
SAMEMOM* 0.00228 0.00143 −0.00736
(8.60) (6.87) (−2.36)
NUMSIC* 0.00745 0.00676 0.00671
(12.39) (12.22) (12.03)
SIZE1* 0.04683 0.04816 0.04855
(11.90) (11.84) (11.66)
SIZE2* 0.01012 0.01021 0.01033
(2.78) (2.79) (2.83)
SIZE1* x SIZE2* −0.06530 −0.06750 −0.06692
(−12.20) (−11.80) (−11.80)
Other controls reported in the Internet Appendix
Nonlinear size controls No Yes Yes Yes

Pair characteristic controls No No Yes Yes
Nonlinear style controls No No No Yes
(Continued)
Table II—Continued
Panel B: Subsample Analysis of Panel A
Dependent Variable: Correlation of Four-Factor Residuals
(1) (2) (3) (4)

First Subsample (1980–1990)
Constant 0.00265 0.00264 0.00158 0.00161

(15.61) (15.39) (5.20) (5.34)
FCAP* 0.00164 0.00135 0.00084 0.00087
(11.81) (12.59) (8.20) (8.77)
Second Subsample (1991–2000)
Constant 0.00493 0.00500 0.00255 0.00256

(5.75) (5.67) (2.85) (2.79)
FCAP* 0.00439 0.00266 0.00182 0.00197
(12.46) (12.06) (9.81) (11.76)
Third Subsample (2001–2008)
Constant 0.00811 0.00815 0.00470 0.00478

(16.88) (16.57) (11.06) (10.34)
FCAP* 0.00607 0.00385 0.00246 0.00281
(13.77) (7.13) (4.56) (5.76)
Panel C: Measuring a Trend in the Time Series of Fama and MacBeth Coefficients
Dependent Variable: Corresponding Coefficient Estimates from Panel A
(1) (2) (3) (4)

Intercept
Constant 0.00508 0.00512 0.00284 0.00288

(15.58) (15.30) (7.92) (7.78)
FCAP* 0.00395 0.00256 0.00168 0.00184
(20.89) (13.58) (9.17) (11.07)
Trend
Constant 0.00008 0.00008 0.00005 0.00005

(9.53) (9.47) (5.62) (5.34)
FCAP* 0.00006 0.00003 0.00002 0.00002
(9.26) (4.55) (2.64) (3.61)
is consistent with analysts choosing to cover stocks that are similar. We also
find a strong sector effect as the coefficient on NUMSIC* is 0.00745 with a
t-statistic of 12.39.
Turning to the style-based controls, we find that stocks with similar mo-
mentum characteristics covary; the coefficient on SAMEMOM* is 0.00228
(t-statistic of 8.60). Similarity in book-to-market has a much smaller effect
as the coefficient on SAMEBM* is 0.00031 (t-statistic of 2.68). Of course, one
should take into account the fact that we are already controlling for exposure
to the book-to-market factor, HML, by forecasting the correlation of four-factor

residuals rather than the correlation of raw returns.
The estimates in the second column also include very general size controls
because of the strong correlation between institutional ownership and size.
These controls include a third-order polynomial in SAMESIZE* as well as a
second-order polynomial in SIZE1*, SIZE2*, and the interaction between the
two. We report the first-order terms of these polynomials in Table II, leaving the
estimates on the higher-order terms for the Internet Appendix.19 Though these
controls are important in describing cross-sectional variation in four-factor
residual correlation, common ownership remains quite significant. In particu-
lar, the estimated coefficient is somewhat smaller (0.00256 from 0.00395) but
remains very statistically significant when we control for style similarity as
well as include nonlinear controls for size.
In the third specification of Table II, Panel A, we further control for a long
list of pair characteristics from Chen, Chen, and Li (2012). These estimates are
reported in the Internet Appendix. All of the Chen, Chen, and Li variables are
statistically significant and enter with the intuitive sign. Nevertheless, FCAP*
remains statistically significant.
The fourth and final column of Table II, Panel A shows results when we allow
not only SAMESIZE* but also SAMEBM* and SAMEMOM* to enter as third-
order polynomials. This regression also incorporates second-order polynomial
functions of the rank-transformed book-to-market and momentum character-
istics of each stock in the pair as well as interaction terms. In this regression,
our most complete and quite flexible specification, common ownership, remains
economically and statistically significant.20
In Table II, Panel B we show that our effect has increased over time. The fact
that the effect becomes stronger is consistent with the significant increase in
the number of active funds over the last 30 years. We first examine subsample
estimates (one for each decade). For the sake of brevity, in each of the three
subsamples we only show the coefficients on Constant and FCAP*, leaving the
estimates on the control variables for the Internet Appendix. When using all
controls, we find that the coefficient on FCAP* moves from a relatively low (but
still statistically significant) 0.00087 in the 1980s, to 0.00197 in the 1990s, and
then to a relatively high estimate of 0.00281 in the 2000s.
Table II, Panel C shows results from an alternative approach to measuring
the evolution of the common ownership effect over time. In particular, we
regress the time series of Fama and MacBeth (1973) coefficients on a constant
and a trend. Again, for brevity, we only show the estimate for the Constant
and FCAP*, relegating estimates for the remaining controls to the Internet
Appendix. In each of the four specifications, we find that the trend coefficient
19 The Internet Appendix may be found in the online version of this article.
20 Although the variation linked to FCAP* may seem modest, note that predicting realized
abnormal return correlation is a difficult task. Realized abnormal return correlation is very noisy,
with an average standard deviation in each cross section of approximately 0.25. R2 s are less than
2% on average, even in the fourth specification, which contains very flexible style controls.
in the specification is positive and statistically significant. For example, the

estimate for specification (4), which includes all controls, is 0.00002 and very
significant with a t-statistic of 3.61.
As a final check of the reasonableness of our method of linking cross-sectional
variation in connectedness to excess comovement, we reestimate the fourth
specification of Table I, Panel A using FCAP instead of FCAP* as well as simple
counts of the number of common owners of the pair (F and F*). These results
can be found in the Internet Appendix. All of our findings continue to hold.
To summarize, the main conclusion from Table II is that common ownership
forecasts statistically and economically significant variation in comovement.21
B. Mutual Fund Scandal: A Natural Experiment

Though the results in the previous section are suggestive, ultimately they
cannot rule out the possibility that mutual fund managers choose to hold sim-
ilar stocks. To confirm that the link we find is causal, we exploit a natural ex-
periment based on the mutual fund scandal that occurred in September 2003.
During that month, 25 fund families settled allegations of illegal trading that
included market timing and late trading. Implicated families had significant
outflows as a consequence of the scandal. Kisin (2011) estimates that funds of
implicated families lost 14.1% of their capital within one year and 24.3% within
two years. These outflows continued until the end of 2006. In stark contrast,
families not implicated increased their capital by nearly 12% over the same
period. We argue that capital flow arising from this scandal is an exogenous
shock, providing a useful way to eliminate concerns that our results are simply
due to the endogenous choice of fund managers.
Our specific instrument is the ratio (RATIO) of the total value held by all
common “scandal” funds of the two stocks over the total value held by all
common funds, as of the time when the scandal broke. To make sure our findings
are robust, we use either RATIO or a dummy that equals one if RATIO is above
median, and present results for both measures.
In the first stage, we regress FCAP* on RATIO, the level of FCAP* as of
September of 2003 (FCAP200309*), and the various sets of control variables
used in regressions (1) to (4) in Table II. We find that the coefficient on RATIO in
these four first-stage regressions, reported in the Internet Appendix, is always
very significant.
Table III presents the results of the second-stage regressions, where we use
the fitted FCAP* from the first stage to predict cross-sectional variation in
four-factor residual correlations. For the sake of brevity, Table III only reports
the coefficient on the instrumented FCAP*. All other coefficient estimates are
reported in the Internet Appendix.
21 We only examine in-sample forecasts of cross-sectional variation in comovement. Since
DeMiguel, Garlappi, and Uppal (2007) argue that 1/N rules perform better out-of-sample than
traditional forecasts of the covariance matrix, it would be interesting to explore how our method
and our characteristics perform out-of-sample. We leave this question for future research.
Table III
Natural Experiment
This table reports results from a 2SLS instrumental variables regression based on a natural exper-
iment. In September 2003, 25 fund families experienced large outflows of capital as a consequence
of a settlement regarding alleged illegal trading. We argue that the outflow of capital because
of this scandal is an exogenous shock unrelated to the endogenous investment decisions of fund
managers. In the first stage we predict the variable FCAP* with the ratio (RATIO) of the total
ownership value held by all common “scandal” funds of the two stocks over the total ownership
value held by all common funds, both measured as of the time of the scandal (end of September
2003). The second stage of the regression uses the fitted FCAP* to forecast the correlation of daily
Fama and French (1993)–Carhart (1997) residuals in month t + 1. Four different specifications are
shown, corresponding to the specifications shown in Table II. We add to each of those four groups
of controls as an additional control for the level of connectedness as of the scandal, FCAP˙200309*.
The first row of estimates directly uses the IV variable, RATIO, in the first stage. The second row
of estimates replaces RATIO in the first stage with a dummy variable that equals one if RATIO
is above its median value. The final row of estimates report the coefficients on FCAP* in an OLS
regression corresponding to Table II, but for this specific subsample (January 2004 to December
2006) and with FCAP˙200309*, for comparison with the IV analysis. We report t-statistics in paren-
theses. The coefficients for the first stage are not shown in this table and are instead reported in
the Internet Appendix. Similarly, estimates on all control variables in the second stage in each of
the 12 specifications are reported in the Internet Appendix.
Dependent Variable: Correlation of Four-Factor Residuals
Specification Instrument (1) (2) (3) (4)
2SLS RATIO 0.04204 0.04261 0.03345 0.02874

(5.39) (3.51) (2.41) (2.03)
2SLS RATIO > Median[RATIO] 0.04107 0.04007 0.03222 0.02826
(5.98) (3.44) (2.50) (2.15)
OLS 0.00181 0.00062 –0.00026 0.00041
(4.43) (1.47) (–0.47) (0.83)

The first row of estimates in Table III corresponds to the 2SLS coefficients on
instrumented FCAP* when the continuous variable RATIO is the instrumen-
tal variable. The coefficient in specification (1), where there are no controls,
is 0.04204, which is highly significant with a t-statistic of 5.39. This causal
effect is economically significant as well, with fitted values ranging from an
average minimum of –0.0019 to an average maximum 0.0184 around an av-
erage abnormal correlation of 0.0089. As one moves across the column to the
right of the table, including more and more controls, the coefficient decreases
somewhat as well as the t-statistic, but remains statistically significant, even
for specification (4), which includes all of our controls.
Similar patterns can be found in the second row of estimates, where a dummy
for above-median RATIO is used as the instrumental variable. The last row of
estimates in Table III shows the coefficients of the OLS regression estimated
in Table II but for the subsample in which the natural experiment takes place
(from December 2003 to December 2006) and with FCAP200309* added as an
additional control, for purposes of comparison. These OLS results are weaker
than the 2SLS results, consistent with the endogeneity of common ownership
being important.
To summarize, Table III shows that the mutual fund scandal generates exoge-
nous variation in common ownership that causes abnormal return correlation
in the following month.
C. When Does Connectedness Matter?

Table II documents that institutional connectedness helps predict cross-
sectional variation in comovement and Table III confirms that the relation
is causal. The remainder of the analysis focuses on exploring why connect-
ing stocks through common fund ownership matters and understanding the
implications of the finding for the cross section of average returns and for
performance attribution of hedge funds.
A likely explanation for our findings is price pressure arising from mutual
fund flows as in Coval and Stafford (2007) and Lou (2012). Indeed, the 2003
mutual fund scandal was chosen as a natural experiment because of the unex-
pected flows the event produced. To provide additional evidence on this inter-
pretation, we exploit cross-sectional heterogeneity in stock pair characteristics.
Specifically, in Table IV we interact FCAP* with two variables, the total float
capitalization of the two stocks in the pair, PFLOAT*, and the absolute value
of the total net flow into the common funds (those that held both i and j at time
t), PFLOW, over the previous quarter.
We follow the literature in defining flows (see Coval and Stafford (2007)).
Accordingly, we measure the net investment flow of funds into fund k in quarter
t as

FLOW k,t = TNAi,t − TNAi,t−1 1 + Ri,t , (2)
where T N Ai,t is the total net assets of fund i in quarter t and Ri,t is the fund
return over the period t − 1 to t reported by CRSP Mutual Fund Database.
Note that TNA is reported quarterly before 1991 and monthly thereafter. To
compute quarterly flows after 1990, we simply sum monthly flows.
Next, for each pair, we compute the total amount of net capital flowing into
the common owners as the absolute value of the sum of FLOW over the K
common funds,

K
PFLOW i, j,t = ABS FLOW k,t . (3)
k=1
We do not rank-transform PFLOW because of the evidence in Coval and

Stafford (2007) that extreme flows are particularly important in understanding
Table IV
Connected Comovement: Cross-Sectional Variation
This table reports Fama and MacBeth (1973) estimates of monthly cross-sectional regressions
forecasting the correlation of daily Fama and French (1993)–Carhart (1997) residuals in month
t + 1 for the sample of stocks defined in Table I. We add interactions between FCAP* and the total
float of the pair (PFLOAT) and the absolute value of the total flows into the common funds holding
the pair (PFLOW). We also include a triple interaction among these three variables. PFLOAT*
is first rank-transformed and then normalized to have unit standard deviation, which we denote
with *. We calculate Newey and West (1987) standard errors (four lags) of the Fama and MacBeth
(1973) estimates that take into account autocorrelation in the cross-sectional slopes. The analysis
is limited to the 1991 to 2008 subperiod when PFLOAT is available. Controls not shown here are
reported in the Internet Appendix.
Dep. Variable: Correlation of Four-Factor Residuals
(1) (2) (3) (4)
Intercept 0.00426 0.00454 0.00084 0.00101

(6.74) (6.51) (1.19) (1.51)
FCAP* 0.00835 0.00428 0.00313 0.00338
(15.95) (12.59) (10.86) (12.17)
FCAP* x PFLOAT* −0.00038 −0.00016 −0.00043 −0.00038
(−1.64) (−0.88) (−2.44) (−2.10)
FCAP* x PFLOW 0.00407 0.00244 0.00253 0.00248
(8.18) (6.17) (7.00) (6.90)
FCAP* x PFLOAT* x PFLOW −0.00051 −0.00028 −0.00050 −0.00046
(−2.15) (−1.39) (−2.58) (−2.41)

price pressure. However, we continue to normalize PFLOW to facilitate the

economic interpretation of the measured coefficients.
Of course, pairs of stocks may differ in their capacity to accommodate the
potential flow. We use the total value of shares available for trading between
the two stocks as a proxy for that capacity. Specifically, we measure PFLOAT
as the total dollar amount of market capitalization available for trading for
the pair of stocks in question. The normalized (to have zero mean and unit
standard deviation) rank-transform of PFLOAT is denoted by PFLOAT*. The
analysis is limited to the 1991 to 2008 subperiod in which PFLOAT is available.
In Table IV we add to the regressions in Table II interaction terms between
FCAP* and the two characteristics described above as well as a triple inter-
action among FCAP*, PFLOAT*, and PFLOW.22 We find that FCAP* more
strongly forecasts comovement for pairs of smaller stocks, as the coefficients on
22 We add the direct effect of PFLOAT* and PFLOW as well as the interaction between PFLOAT*
and PFLOW to the control variables to ensure that any triple interaction effect we find is not simply
because of an omitted variable. For the sake of brevity, the table only reports the coefficients on
FCAP* and its interactions. All other coefficient estimates are reported in the Internet Appendix.
the interaction of FCAP* with PFLOAT* are negative, though not statistically
significant in all four specifications. Similarly, FCAP* more strongly forecasts
comovement when PFLOW is high, that is, when total flows to common funds
are very high or very low. This interaction is always very statistically sig-
nificant. Finally, the triple interaction reveals that the link between common
ownership and future comovement is strongest when common owners of a pair
of low-float stocks experience extreme flows. This interaction is statistically
significant in three of the four specifications.
The predicted variation in abnormal return correlation conditional on these
two pair-based characteristics is now of course larger than in Table II. In
specification (1), for example, fitted values linked to FCAP* now range from an
average minimum of –0.0462 to an average maximum 0.0291.
These results are consistent with the causal interpretation provided by the
natural experiment: prices are more subject to nonfundamental comovement
when pressure from net mutual fund trading is high and especially so when
the stocks in question have low float and thus are less liquid.
The Internet Appendix further analyzes the nature of this comovement using
the Campbell (1991) return decomposition, as applied in Campbell, Polk, and
Vuolteenaho (2010).23 The results presented in the Internet Appendix show
that the majority of the variation in comovement linked to connectedness arises
from the covariance of cash flow news and discount rate news. This finding is
consistent with our finding of a causal relation between connectedness and
comovement. News about fundamentals in one stock (cash-flow news) results
in temporary movements (discount-rate news) in another stock as connected
owners sell off positions in both stocks.24
D. Connected-Stock Trading Strategies

If common ownership causes comovement through price pressure, then we
can link ownership-based connectedness to predictable variation in returns.
Specifically, if stock i experiences a negative shock and connected stock j’s price
also drops, we conjecture that stock i’s drop is due to price pressure, which we
expect to revert. Our trading strategy is thus very simple: we buy (sell) stocks
that have gone down (up) if their connected stocks have gone down (up) as
well. This strategy uses the connected return as a signal of whether stock i
is undervalued or overvalued, and we interpret such a strategy as exploiting
the price pressure induced by common ownership.25 Moreover, since one would
23 Of course, these conclusions depend on having correctly specified the VAR producing the
return decomposition. The Internet Appendix describes the methodology and presents the VAR
estimates.
24 Alternatively, we could have found that the majority of the variation in comovement linked
to connectedness is entirely because of cash-flow news covariance. If so, this finding would be
inconsistent with institutional ownership causing comovement as stock owners cannot directly
affect cash flows.
25 Chen, Chen, and Li (2012) also measure the cross-sectional variation in pair-wise correlations
and show that a large portion of that cross-sectional variation is persistent, yet unexplained by a
expect forced sells to be more likely than forced buys, we conjecture that most
of the predictable variation will come from the long side of the strategy.
Since our ownership data are updated quarterly, we look for these patterns
using past three-month returns. By using the actual connected return as a mea-
sure of the price impact from mutual fund trading, we avoid having to measure
the impact of flows on stock returns. In particular, each month we sort our
subset of big stocks into quintiles based on their past three-month return.
We independently sort these stocks into quintiles based on the past three-
month return on the portfolio of stocks that they are connected to through
common ownership (riC,t ). We exploit the results of Table II to construct the
connected portfolio and isolate the incremental effect of common ownership as
well as possible. Specifically, we first measure the degree of abnormal connec-
tions (FCAPO*) by orthogonalizing FCAP* each period to the full set of controls
of regression (4) in Table II.26 We then use FC AP Oi∗j,t to generate the weights
determining riC,t . Specifically, we define
FCAPi∗∗j,t = rank(FCAPOi∗j,t ) if FCAPi j,t > 0 (4)
FCAPi∗∗j,t = 0 if FCAPi j,t = 0. (5)
The return on the portfolio is then

J ∗∗
j=1 FC APi j,t−1 r j,t
riC,t = J . (6)
∗∗
j=1 FC APi j,t−1
We first examine the buy-and-hold abnormal returns on an equal-weighted

portfolio of stocks that are in the low own-return and low connected-return
portfolio (which we dub the low portfolio) and an equal-weighted portfolio of
stocks that are in the high own-return and high connected-return portfolio (the
high portfolio). Specifically, we generate the cumulative buy-and-hold abnormal
return by regressing the t + 1, t + 2, . . . , t + 12 returns on the corresponding
returns on five factors: the four factors of Fama and French (1993) and Carhart
(1997), augmented with the short-term reversal factor (STR) that makes a
size-stratified bet on the one-month reversal effect.27 We include STR as we
are sorting stocks on their past return, and we want to make sure that our
connected-stock strategy is distinct from that effect.28
Figure 1 graphs the cumulative abnormal returns on the low portfolio, the
high portfolio, and a portfolio that is long low and short high. The patterns
long list of variables. They do not incorporate holdings data in their forecasts of correlation. Chen,
Chen, and Li also develop a trading strategy. However, their focus is on enhancing the profitability
of pairs-based trading strategies that look for divergence in the prices of a pair of similar stocks.
The strategy then profits from the correction of the divergence. In contrast, our strategy identifies
stocks that temporarily move together and profits from their eventual divergence.
26 We thank an anonymous referee for this suggestion.
27 The short-term reversal factor is also from Ken French’s website.
28 We find that including STR in our benchmarking has very little effect on our results.
5%
Low minus High

4%
Low
3% High
2%
1%
0%
0 1 2 3 4 5 6 7 8 9 10 11 12
-1%
-2%
Months aŌer porƞolio formaƟon
Figure 1. Cumulative alphas for a connected-stock strategy. This figure graphs the abnor-
mal buy-and-hold performance of a cross-stock-reversal trading strategy that exploits information
in ownership connections among big stocks (stocks above the median NYSE market capitaliza-
tion). These stocks are sorted into 25 portfolios based on independent quintile sorts on their own
three-month return and the three-month return on their connected stock portfolio. We define a
stock’s connected portfolio as those stocks each quarter that have active mutual fund owners in
common with the stock in question. The connected portfolio weights are proportional to the degree
of abnormal common fund ownership, as defined in the text. The figure plots the buy-and-hold ab-
normal returns on stocks that are in the low own-return and low connected-return portfolio (which
we dub the low portfolio), stocks that are in the high own-return and high connected-return port-
folio (the high portfolio), and the difference between low and high abnormal returns. Returns are
benchmarked against the Fama and French (1993)–Carhart (1997) four-factor model augmented
with the short-term reversal factor.
shown in the graph are consistent with stocks being pushed away from funda-
mental value by mutual fund trading, with the connected return being a useful
measure of the extent of that temporary misvaluation. These patterns last for
the next six months. Thus, compared to the standard short-term reversal effect,
the misvaluation is larger and takes more time to revert. Consistent with the
forced selling/price pressure of Coval and Stafford (2007), we find that more of
the effect comes from the low portfolio than the high portfolio.
As a consequence, we evaluate the average returns on “composite” portfolios
that take these predictable patterns in the cross section of average returns into
account. To ensure our findings are not due to the one-month reversal effect,
we first skip a month after the sort and then hold the stocks in question for five
months, following the methodology of Jegadeesh and Titman (1993). Table V
reports the five-factor alphas on these 25 composite portfolios, each of which
Table V
Alphas on Connected-Stock Trading Strategies
This table presents the profitability of a simple trading strategy exploiting stock connectedness.
We independently sort stocks into quintiles based on their own return over the last three months
and the return on their connected portfolio over the last three months. We first measure the
degree of abnormal connections by orthogonalizing FCAP* (which we denote by FCAPO*) each
period to the full set of controls in regression (4) of Table II. We then define the connected return

as riC,t = Jj=1 FCAPi∗∗j,t−1 r j,t / Jj=1 FCAPi∗∗j,t−1 , where FCAPi∗∗j,t = rank(FCAPOi∗j,t ) if FCAPi j,t > 0
∗∗
and FCAPi j,t = 0 if FC APi j,t = 0. Following Jegadeesh and Titman (1993), each composite port-
folio below is an equal-weighted average of the corresponding simple strategies initiated one to
five months prior. The table reports the five-factor alphas on these 25 composite portfolios. The
five factors include the four Fama and French (1993)–Carhart (1997) factors plus a short-term
reversal factor, all downloaded from Ken French’s website. We also report the average returns
on a connected-stock trading strategy (CS) that buys the low own-return low connected-return
composite portfolio and sells the high own-return high connected-return) composite portfolio.
Five factor alphas
Connected portfolio return
Low 2 3 4 High L–H Avg L – H
Low 0.0055 0.0039 0.0039 0.0028 0.0002 0.0053

(4.58) (3.73) (3.79) (2.56) (0.15) (3.57)
2 0.0047 0.0041 0.0027 0.0024 0.0008 0.0039
Own (4.59) (4.73) (3.29) (2.55) (0.84) (3.68)
Return 3 0.0031 0.0024 0.0014 0.0011 −0.0005 0.0036 0.0036
(3.33) (2.87) (1.58) (1.25) (−0.56) (3.56) (4.13)
4 0.0019 0.0007 0.0002 −0.0004 −0.0011 0.0030
(2.10) (0.78) (0.31) (−0.47) (−1.30) (3.18)
High 0.0003 0.0002 −0.0017 −0.0017 −0.0022 0.0025
(0.27) (0.30) (−2.00) (−2.00) (−2.40) (1.90)
L–H 0.0052 0.0036 0.0056 0.0045 0.0023 0.0076 LL–HH (CS)

(3.74) (3.08) (4.48) (3.54) (1.78) (4.96)
is an equal-weighted average of the corresponding simple strategies initiated

one to five months prior.
Two general patterns in Table V are consistent with our initial conclusions
concerning Figure 1. Alphas on the composite portfolios generally increase as
one moves from high to low connected returns within an own-return quintile.
Within each connected-return quintile, the alphas on these composite portfo-
lios generally increase as one moves from high to low own-return portfolios.
These patterns confirm the information in connected returns about a stock’s
subsequent return; as a consequence, we design a long-short trading strategy
to exploit that information. Our connected-stock strategy, which we denote by
CS, buys the composite low portfolio and sells the composite high portfolio. The
five-factor alpha for CS is an impressive 76 basis points per month (t-statistic
of 4.96).29 Consistent with our forced-trading story, more than 71% of the alpha
comes from the composite low portfolio.
For comparison, Table V also reports the average five-factor alpha on a
strategy that instead buys the average (across the own-return quintiles) low
connected-return composite portfolio and sells the average (across the own-
return quintiles) high connected-return portfolio. This strategy earns 36 basis
points per month (t-statistic of 4.13).30 Since such a strategy ignores the in-
formation in the interaction between a stock’s own return and its connected
return, the strong performance further confirms that our finding of a success-
ful connected-stock trading strategy is distinct from the short-term reversal
effect.
Table VI reports the regression loadings from the performance attribution for
the connected-stock trading strategy as well as the effect of including additional
variables in the attribution regression. We first include the liquidity factor of
Pástor and Stambaugh (2003) to measure how CS covaries with this factor.31
We find that CS covaries positively with their factor although the coefficient is
only marginally significant. We then include a trend in the regression. We find
no evidence that the alpha of our connected-stock trading strategy has been
trending over time. Finally, we include dummy variables for each quarter. We
find that our effect is strongest in the fourth quarter of the year but that alphas
in each of the four quarters remain economically and statistically significant.
E. Hedge Fund Index Attribution

Our last analysis uses our connected-stocks trading strategy to analyze the
performance of returns on two CSFB/Tremont Hedge Fund Indexes. These
indexes have been used in a number of studies including Asness, Krail, and
Liew (2001), Agarwal and Naik (2004), Getmansky, Lo, and Makarov (2004),
and Bondarenko (2004). The first index we study is the index of all hedge
funds. As CFSB weights hedge fund returns by assets under management and
captures more than 85% of all assets under management in this investing space,
this index gives a good representation of the extent to which the general health
of the hedge fund industry can be linked to our connected-stocks strategy.32 We
also examine the performance of the long-short component of the CSFB index
29 If we exclude STR from the factor regression, the alpha is 78 basis points and the corre-
sponding t-statistic is 5.10. When we use FCAP* instead of FCAPO* to determine the weights
on a stock’s connected portfolio, the results (reported in the Internet Appendix) are weaker but
are still economically large, generating 48 basis points per month with a corresponding t-statistic
of 2.76. This finding shows the robustness of our result, as well as the usefulness of measuring
abnormal common ownership. We also examined different lengths of the formation period in which
we measure the connected and own returns; our results are robust to formation periods ranging
from one month (45 basis points, t-statistic 3.77) to 12 months (32 basis points, t-statistic 1.56).
30 The FCAP*-weighted version of this strategy earns 30 basis points per month (t-statistic of
2.72).
31 Since this factor is not a traded portfolio, the loading of CS on the liquidity factor is the only
object of interest in this regression.

32 Note that the CFSB does not include managed accounts or funds of funds in its indexes.
Table VI
The Connected-Stock Trading Strategy and Liquidity Risk
This table measures the loadings of the connected-stock trading strategy on a liquidity factor as
well as on trend and quarter dummies. We study the connected strategy, CS, formed in Table V,
which buys the low own-return low connected-return) composite portfolio and sells the high own-
return high connected-return composite portfolio. We regress the return on this trading strategy
on a constant, the liquidity factor (PS INNOV) from the work of Pástor and Stambaugh (2003), the
Fama and French (1993)–Carhart (1997) factors, a short-term reversal factor (STR), a trend, and
seasonal (quarterly) dummies over the period June 1980 to December 2008. Column (1) reports
loadings of the return on our connected strategy on the five factors used in Table V. Column (2) adds
PS INNOV as an explanatory variable. Columns (3) and (4) include a trend or quarterly seasonal
dummies, respectively, as additional explanatory variables. We report t-statistics in parentheses.
Dependent Variable: Connected-Stock Trading Strategy Return
(1) (2) (3) (4)
Alpha 0.0076 0.0076 0.0076 0.0147

(4.96) (4.99) (4.94) (5.08)
PS INNOV 0.0433
(1.73)
RMRF −0.0623 −0.0832 −0.0597 −0.0654
(−1.68) (−2.14) (−1.60) (−1.77)
SMB −0.2828 −0.2861 −0.2831 −0.2695
(−5.73) (−5.81) (−5.73) (−5.44)
HML −0.1502 −0.1595 −0.1469 −0.1343
(−2.77) (−2.93) (−2.69) (−2.47)
UMD −0.8565 −0.8545 −0.8561 −0.8584
(−23.40) (−23.40) (−23.30) (−23.60)
STR 0.0327 0.0368 0.0329 0.0375
(0.70) (0.79) (0.70) (0.81)
Trend 0.0000
(0.61)
Q1 −0.0071
(−1.71)
Q2 −0.0132
(−3.22)
Q3 −0.0086
(−2.09)
Obs 341 341 341 341
R2 67% 68% 67% 68%
to measure the extent to which funds that specifically invest in equities are
exposed to the connected stocks factor.
Table VII reports the results of this analysis. We find that hedge funds in gen-
eral and long-short managers in particular load negatively on the connected-
stock trading strategy. The coefficient in the first column of Table VII estimates
a regression of the overall hedge fund index excess return on the return on our
connected strategy and the four factors of Fama and French (1993) and Carhart
(1997), augmented with the short-term reversal factor, STR. We include the in-
teraction of CS with the lagged value of the change in the VIX index (VIX) to
capture any time-variation in the sensitivity to CS. Both theory and evidence
Table VII
Hedge Fund and Mutual Fund Exposure to Connected Strategy
This table measures the exposure of two CSFB hedge fund return indexes (all and long-short)
as well as the value-weighted average active mutual fund return (net of fees) to the connected
strategy described in Table V. We regress fund index returns in excess of the Treasury bill rate
on a constant, the connected strategy, and either the eight Fung and Hsieh (2001, 2004) hedge
fund factors or the Fama and French (1993)–Carhart (1997) model plus a short-term reversal
factor (STR). The time period is January 1994 to December 2008. The analysis adds as additional
explanatory variables the return on the connected-stock trading strategy (CS) of Table V as well
as an interaction between CS and the lagged, demeaned, normalized change in the VIX (VIX).
We report t-statistics in parentheses.
HF All HF Long/Short MF All
Alpha 0.0025 0.0030 0.0029 0.0027 −0.0009

(1.93) (2.63) (2.91) (2.42) (−2.45)
CS −0.0490 −0.1306 −0.0989 −0.2245 −0.0225
(−1.13) (−5.38) (−2.88) (−9.25) (−1.78)
CS * VIXt−1 −0.0070 −0.0034 −0.0140 −0.0124 −0.0018
(−1.30) (−0.67) (−3.29) (−2.49) (−1.18)
RMRF 0.3328 0.4658 0.9518
(10.85) (19.17) (106.5)
SMB 0.0794 0.1345 0.0553
(2.02) (4.33) (4.84)
HML 0.0236 −0.1073 −0.0273
(0.58) (−3.35) (−2.31)
UMD 0.1026 0.1353 0.0057
(2.28) (3.80) (0.44)
STR −0.0428 −0.0580 −0.0189
(−1.30) (−2.23) (−1.97)
Bond Trend −0.0182 −0.0041
(−2.28) (−0.51)
Currency Trend 0.0131 0.0066
(2.09) (1.07)
Commodity Trend 0.0112 −0.0011
(1.26) (−0.13)
Equity Market 0.1578 0.3423
(4.12) (8.96)
Size Spread 0.0499 0.1825
(1.39) (5.10)
Bond Market −0.1531 −0.0551
(−4.10) (−1.48)
Credit Spread −0.2298 −0.0669
(−4.27) (−1.25)
Emerging Market 0.0903 0.0929
(3.61) (3.72)
Obs 182 182 182 182 182
R2 52% 59% 82% 75% 99%
suggest that an increase in the VIX is a good measure of turbulent times for the
hedge fund industry.33 We demean and standardize VIX so that the coefficient
on CS reflects the return loading for typical values of VIX, and the interaction
coefficient indicates how a one-standard-deviation move affects that loading.
The coefficients on CS and CS interacted with VIX for the all-hedge-fund
index are negative but statistically insignificant in a six-factor regression that
adds CS to the five factors of Table VI. The second column of the table instead
attributes the performance of the hedge fund index to the connected strategy,
its interaction with VIX, and the eight hedge fund factors of Fung and Hsieh
(2001, 2004).34 Though hedge funds in aggregate load on these eight factors
to various degrees, our connected-stock factor is important in describing the
returns on hedge funds. The coefficient is now more economically and statis-
tically significant: the point estimate for CS is –0.1306 and has an associated
t-statistic of –5.38. The interaction term remains statistically insignificant.
This result suggests that our trading strategy is a useful tool to measure the
state of the hedge fund industry as the typical hedge fund is negatively exposed
to our factor, regardless of whether the VIX is increasing or decreasing.
Perhaps more interesting results are in the third and fourth columns of
Table VII. In column (3), we measure the degree to which the long-short subset
of hedge funds covaries with our connected-return trading strategy in the pres-
ence of the Fama–French–Carhart factors and the short-term reversal factor.
In column (4), we instead use the Fung and Hsieh factors as controls. In both
cases, we find that the returns on this subset of hedge funds strongly and nega-
tively covary with our connected-return factor and particularly so in turbulent
times (when VIX is increasing).
Specifically, we find that the average sensitivity to CS for long-short hedge
funds is more than 70% larger in absolute value than the corresponding esti-
mate for all hedge funds. The t-statistics are correspondingly larger and indi-
cate strong statistical significance (a t-statistic of –2.88 when controlling for the
five equity factors and –9.25 when controlling for the eight Fung and Hsieh fac-
tors). The magnitude of the interaction coefficients is more than twice as large
for the long-short fund subset compared to all hedge funds and t-statistics in
the attribution regressions are –3.29 and –2.49, respectively. These findings are
comforting as one would expect this subset of hedge funds to be more exposed
to our factor on average and particularly so when the VIX is increasing.
For the sake of comparison, we also estimate the loading of a value-weighted
portfolio consisting of all of the active mutual funds in our sample over the
same time period. This portfolio has a smaller (in absolute value) sensitivity to
the connected strategy as the estimate is –0.0225 with an associated t-statistic
of –1.78. Though we do not observe complete holdings data for all hedge funds
33 Gromb and Vayanos (2002) and Brunnermeier and Pedersen (2009) predict that increasing
volatility tightens the funding constraints of liquidity providers like hedge funds. Indeed, hedge
funds face redemptions and lower leverage when the VIX is increasing (Ang, Gorovyy, and van
Inwegen (2011) and Ben-David, Franzoni, and Moussawi (2012)).
34 We downloaded three of the Fung and Hsieh (2001) factors from http://faculty.
fuqua.duke.edu/˜dah7/DataLibrary/TF-FAC.xls.
0.06 0.00
All-hedge-fund loading on CS (left axis)

0.04 -0.05
Long-short hedge fund loading on CS (left axis)
Cumulative abnormal performance of CS (right axis)
0.02 -0.10
0.00 -0.15
-11 -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12
-0.02 -0.20
-0.04 -0.25
-0.06 -0.30
-0.08 -0.35
-0.10 -0.40
-0.12 -0.45
-0.14 -0.50
Figure 2. Hedge fund sensitivity to the connected-stocks trading strategy return. This
figure plots the loadings of hedge fund index returns on the connected stocks trading strategy,
CS, in event time as well as the corresponding cumulative five-factor abnormal return on the CS
strategy.
and therefore cannot see the exact positions of these long-short hedge funds,
these results suggest that these hedge funds do not take full advantage of the
opportunities that price pressure from mutual fund flows provide. Indeed, these
results are consistent with hedge funds exacerbating rather than mitigating
the price pressure patterns documented in this paper.35
Figure 2 suggests why it is not surprising that the typical hedge fund loads
negatively on our connected strategy. This figure plots both the loadings of the
two hedge fund indexes on the connected strategy as well as the cumulative
abnormal return on the connected strategy in event time, where the event is the
forming of the connected-stock trading strategy. One reasonable interpretation
of this figure is that hedge funds follow a momentum strategy that effectively
front-runs mutual fund flows, but the typical hedge fund is unable to exit
its positions in time and therefore exacerbates the price dislocation it helps
initiate.
Of course, this finding begs the question as to whether it is value-adding
for long-short hedge funds to follow a front-running strategy. In Table VII, the
35 In the Internet Appendix, we repeat the analysis when the strategy is computed using FCAP*
instead of FCAPO*. We find results that are qualitatively similar.

unconditional loading on CS is –0.0989 while the corresponding loading on

MOM is 0.1353. Moreover, the loading on CS decreases by 0.0140 for every
unit change in the normalized VIX. Since we find that MOM has an alpha
(controlling for the other factors in Table VII, including CS) of 71 basis points,
the total incremental abnormal return for this joint exposure to MOM and CS is
small but remains positive unless the normalized change in VIX is nearly two.
Of course, this analysis ignores any potential time-variation in the sensitivity
of hedge funds to MOM. Thus, the extent to which hedge funds can effectively
front-run the anticipated forced trades of mutual funds remains an interesting
question for future research.36
IV. Conclusion
We show that stocks are connected through the mutual fund owners they
have in common. In particular, pairs of stocks that are connected in this fashion
covary more together, controlling for exposure to systematic return factors,
style and sector similarity, and many other pair characteristics. Evidence from
a natural experiment—the 2003 mutual fund trading scandal—confirms that
this relation is causal. Consistent with these findings, we further show that
the general link between shared ownership and comovement is stronger for
owners who are experiencing extreme flows in low-float stocks.
These results motivate a novel cross-stock-reversal trading strategy that ex-
ploits the information in ownership connections to generate annual abnormal
returns of more than 9%. As a consequence, we provide a simple way to docu-
ment the extent to which ownership-based connections result in equity market
contagion. In an application, we document that the typical long-short hedge
fund covaries negatively with our trading strategy (and more so than the typ-
ical mutual fund we initially study), suggesting that hedge funds on average
may be part of the cause of the excess covariation and price dislocation that
contagion from ownership-based connections generates.
Initial submission: April 1, 2010; Final version received: November 12, 2013
Editor: Campbell Harvey
REFERENCES
Agarwal, Vikas, and Narayan Y. Naik, 2004, Risk and portfolio decisions involving hedge funds,
Review of Financial Studies 17, 63–98.
Ang, Andrew, Sergiy Gorovyy, and Gregory B. van Inwegen, 2011, Hedge fund leverage, Journal
of Financial Economics 102, 102–126.
Antón, Miguel, 2011, Cash-flow driven covariation, Working paper, IESE Business School.
Asness, Clifford, Robert Krail, and John Liew, 2001, Do hedge funds hedge?, Journal of Portfolio
Management 28, 6–19.
36 In somewhat related work, Lou and Polk (2013) find that the typical long-short equity hedge
fund is able to time the momentum factor, but this timing ability is decreasing in the size of the
fund’s assets under management.
Barberis, Nicholas, and Andrei Shleifer, 2003, Style investing, Journal of Financial Economics 68,
161–199.
Barberis, Nicholas, Andrei Shleifer, and Jeffrey Wurgler, 2005, Comovement, Journal of Financial
Economics 75, 283–317.
Bartram, Söhnke, John Griffen, and David Ng, 2012, How important are foreign ownership link-
ages for international stock returns, Working paper, University of Texas.
Bekaert, Geert, Campbell R. Harvey, and Angela Ng, 2005, Market integration and contagion,
Journal of Business 78, 39–79.
Ben-David, Itzhak, Francesco Franzoni, and Rabih Moussawi, 2012, Hedge fund stock trading in
the financial crisis of 2007–2008, Review of Financial Studies 25, 1–54.
Bondarenko, Oleg, 2004, Market price of variance risk and performance of hedge funds, Working
paper, University of Illinois, Chicago.
Brunnermeier, Markus, and Lasse Pedersen, 2009, Market liquidity and funding liquidity, Review
of Financial Studies 22, 2201–2238.
Campbell, John Y., 1991, A variance decomposition for stock returns, Economic Journal 101, 157–
179.
Campbell, John Y., Christopher Polk, and Tuomo Vuolteenaho, 2010, Growth or glamour? Funda-
mentals and systematic risk in stock returns, Review of Financial Studies 23, 305–344.
Carhart, Mark, 1997, On persistence in mutual fund performance, Journal of Finance 52, 56–82.
Chen, Huafeng, Shaojun Chen, and Feng Li, 2012, Empirical investigation of an equity pairs
trading strategy, Working paper, University of British Columbia.
Chen, Joseph, Samuel Hanson, Harrison Hong, and Jeremy C. Stein, 2008, Do hedge funds profit
from mutual-fund distress?, Working paper, Harvard University.
Chevalier, Judith, and Glen Ellison, 1997, Risk taking by mutual funds as a response to incentives,
Journal of Political Economy 105, 1167–1200.
Coval, Josh, and Eric Stafford, 2007, Asset fire sales (and purchases) in equity markets, Journal
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.
Daniel, Kent, and Sheridan Titman, 1997. Evidence on the characteristics of cross sectional vari-
ation in stock returns, Journal of Finance 52, 1–33.
DeMiguel, Victor, Lorenzo Garlappi, and Raman Uppal, 2007, Optimal versus naive diversification:
How inefficient is the 1/N portfolio strategy? Review of Financial Studies 22, 1915–1953.
Duffie, Darrell, 2010, Asset price dynamics with slow-moving capital, Journal of Finance 65, 1238–
1268.
Ellul, Andrew, Pab Jotikasthira, and Christian T. Lundblad, 2010, Regulatory pressure and fire
sales in the corporate bond market, Journal of Financial Economics 101, 596–620.
Fama, Eugene F., and Kenneth R. French, 1993, Common risk factors in the returns on stocks and
bonds, Journal of Financial Economics 33, 3–56.
Fama, Eugene F., and Kenneth R. French, 2008, Dissecting anomalies, Journal of Finance 63,
1653–1678.
Fama, Eugene, and James D. MacBeth, 1973, Risk, return, and equilibrium: Empirical tests,
Journal of Political Economy 81, 607–636.
Forbes, Kristin, and Roberto Rigobon, 2001, No contagion, only interdependence: Measuring stock
market co-movements, Journal of Finance 57, 2223–2261.
Fung, William, and David A. Hsieh, 2001, The risk in hedge fund strategies: Theory and evidence
from trend followers, Review of Financial Studies 14, 313–341.
Fung, William, and David A. Hsieh, 2004, Hedge fund benchmarks: A risk-based approach, Finan-
cial Analysts Journal 60, 65–80.
Getmansky, Mila, Andrew W. Lo, and Igor Makarov, 2004, An econometric model of serial correla-
tion and illiquidity in hedge fund returns, Journal of Financial Economics 74, 529–609.
Greenwood, Robin, and David Thesmar, 2011, Stock price fragility, Journal of Financial Economics
102, 471–490.
Gromb, Denis, and Dimitri Vayanos, 2002, Equilibrium and welfare in markets with financially
constrained arbitrageurs, Journal of Financial Economics 66, 361–407.
Gromb, Denis, and Dimitri Vayanos, 2010, Limits of arbitrage: State of the theory, Annual Review
Ippolito, Richard A., 1992, Consumer reaction to measures of poor quality: Evidence from the
mutual fund industry, Journal of Law and Economics 35, 45–70.
Jegadeesh, Narasimhan, and Sheridan Titman, 1993, Returns to buying and selling losers: Impli-
cations for stock market efficiency, Journal of Finance 48, 65–91.
Jotikasthira, Chotibhak, Christian Lundblad, and Tarun Ramadorai, 2012, Asset fire sales and
purchases and the international transmission of funding shocks, Journal of Finance 67 2015–
2050.
King, Mervyn, Enrique Sentana, and Sushil Wadhwani, 1994, Volatility and links between national
stock markets, Econometrica 62, 901–933.
King, Mervyn, and Sushil Wadhwani, 1990, Transmission of volatility between stock markets,
Kisin, Roni, 2011, The impact of mutual fund ownership on corporate investment: Evidence from
a natural experiment, Working paper, Washington University of St. Louis.
Lou, Dong, 2012, A flow-based explanation for return predictability, Review of Financial Studies
25, 3457–3489.
Lou, Dong, and Christopher Polk, 2013, Comomentum: Inferring arbitrage activity from return
correlations, Working paper, London School of Economics.
Mitchell, Mark, Lasse Heje Pedersen, and Todd Pulvino, 2007, Slow moving capital, American
Economic Review, 97, 215–220.
Newey, Whitney K., and Kenneth D. West, 1987, A simple, positive semi-definite, heteroskedastic-
ity and autocorrelation consistent covariance matrix, Econometrica 55, 703–708.
Pástor, Ľuboš, and Robert Stambaugh, 2003, Liquidity risk and expected stock returns, Journal
of Political Economy 111, 642–685.
Rigobon, Roberto, 2002, Contagion: How to measure it? in Sebastian Edwards and Jeffrey
Frankel eds.: Preventing Currency Crises in Emerging Markets (The University Chicago Press,
Chicago).
Sadka, Ronnie, 2010, Liquidity risk and the cross-section of hedge-fund returns, Journal of Finan-
cial Economics 98, 54–71.
Scholes, Myron, 1972, The market for corporate securities: Substitution versus price pressure and
the effects of information on stock prices, Journal of Business 45, 179–211.
Shleifer, Andrei, and Robert Vishny, 1992, Liquidation values and debt capacity: A market equi-
librium approach, Journal of Finance 47, 1343–1366.
Shleifer, Andrei, and Robert Vishny, 1997, The limits to arbitrage, Journal of Finance 52, 35–55.
Singleton, Kenneth J., 2011, Investor flows and the 2008 boom/bust in oil prices, Working paper,
Stanford University.
Sirri, Eric, and Peter Tufano, 1998, Costly search and mutual fund flows, Journal of Finance 53,
1589–1622.
Sun, Zheng, 2007, Clustered institutional holdings and stock comovement, Working paper, New
York University.
Vayanos, Dimitri, 2004, Flight to quality, flight to liquidity, and the pricing of risk, Working paper,
London School of Economics.
Vayanos, Dimitri, and Paul Woolley, 2013, An institutional theory of momentum and reversal,
Vijh, Anand M., 1994, S&P 500 trading strategies and stock betas, Review of Financial Studies 7,
215–251.
Supporting Information
Additional Supporting Information may be found in the online version of this
article at the publisher’s web site:
Appendix S1: Internet Appendix.

Anton JF 2014

Uploaded by

Copyright:

Available Formats

Anton JF 2014

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Anton JF 2014

Uploaded by

Copyright:

Available Formats

THE JOURNAL OF FINANCE • VOL. LXIX, NO.

MIGUEL ANTÓN and CHRISTOPHER POLK∗

return on stocks connected to a particular ﬁrm to predict cross-sectional varia-

II. Data and Methodology

B. Measuring Common Ownership

example, Fama and French (2008).

C. Modeling Cross-Sectional Variation in Comovement

However, since style characteristics do not line up perfectly with regression

are not issued as consistently.

Panel A: Number of Stocks, Pairs, and Funds

Year Stocks Pairs Funds

1980 895 400,065 171

Panel B: Summary Statistics for Stocks and Funds

Variable Period Mean Median SD Min Max

Num stocks per fund 1980–2008 58.5 41 69.8 1 1028

Num funds per stock 1980–2008 68.6 39 84.1 1 829

Panel C: The Cross-Sectional Distribution of Common Fund Ownership

Year Mean SD 0% 25% 50% 75% 95% 99% 100%

Panel A: Full Sample (1980–2008)

Dependent Variable: Correlation of 4F Residuals

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

Constant 0.00508 0.00512 0.00284 0.00288

Nonlinear size controls No Yes Yes Yes

Panel B: Subsample Analysis of Panel A

Dependent Variable: Correlation of Four-Factor Residuals

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

Constant 0.00265 0.00264 0.00158 0.00161

Constant 0.00493 0.00500 0.00255 0.00256

Constant 0.00811 0.00815 0.00470 0.00478

Dependent Variable: Corresponding Coefﬁcient Estimates from Panel A

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

Constant 0.00508 0.00512 0.00284 0.00288

Constant 0.00008 0.00008 0.00005 0.00005

to the book-to-market factor, HML, by forecasting the correlation of four-factor

in the speciﬁcation is positive and statistically signiﬁcant. For example, the

B. Mutual Fund Scandal: A Natural Experiment

21 We only examine in-sample forecasts of cross-sectional variation in comovement. Since

Dependent Variable: Correlation of Four-Factor Residuals

Speciﬁcation Instrument (1) (2) (3) (4)

2SLS RATIO 0.04204 0.04261 0.03345 0.02874

Nonlinear size controls No Yes Yes Yes

C. When Does Connectedness Matter?

We do not rank-transform PFLOW because of the evidence in Coval and

Dep. Variable: Correlation of Four-Factor Residuals

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

Intercept 0.00426 0.00454 0.00084 0.00101

Nonlinear size controls No Yes Yes Yes

price pressure. However, we continue to normalize PFLOW to facilitate the

D. Connected-Stock Trading Strategies

FCAPi∗∗j,t = rank(FCAPOi∗j,t ) if FCAPi j,t > 0 (4)

FCAPi∗∗j,t = 0 if FCAPi j,t = 0. (5)

The return on the portfolio is then

We ﬁrst examine the buy-and-hold abnormal returns on an equal-weighted

Low minus High

Five factor alphas

Connected portfolio return

Low 2 3 4 High L–H Avg L – H

Low 0.0055 0.0039 0.0039 0.0028 0.0002 0.0053

L–H 0.0052 0.0036 0.0056 0.0045 0.0023 0.0076 LL–HH (CS)

is an equal-weighted average of the corresponding simple strategies initiated