The Role of Equity Funds

in the Financial Crisis Propagation

Harald Hau
University of Geneva and Swiss Finance Institute

Sandy Lai
University of Hong Kong
Abstract
The early stage of the 200708 financial crisis was marked by large value losses
for bank stocks. This paper identifies the equity funds most aected by this valuation shock and examines its consequences for the nonfinancial stocks owned by
the respective funds. We document three key empirical findings. First, ownership
links to these "distressed equity funds" lead to large underperformance of the most
exposed nonfinancial stocks. The results hold even after controlling for various accounting valuation measures. Second, distressed fund sales and the associated price
discounts are concentrated among those exposed stocks that perform relatively well.
Third, stocks with higher overall fund ownership generally performed much better
throughout the crisis.

JEL Classification: G11, G14, G23

Keywords: Financial Crisis, Crisis Spillover Eect, Mutual Fund Ownership

University of Geneva and Swiss Finance Institute, Unipignon, 40 Bd du Pont dArve, CH - 1211 Geneva 4,
Switzerland. Telephone: (++41) 22 379 9581. E-mail: [email protected]. Web page: http://www.haraldhau.com.

School of Economics and Finance, University of Hong Kong, K.K. Leung Building, Pokfulam Road, Hong
Kong. Telephone: (++852) 2219-4180. E-mail: [email protected]. Web page: http://www.sandylai-research.com.
We acknowledge the generous support of Seth Payne from NYSE Technologies Global Market Data, who
provided us with the retail trading volume data on NYSE listed stocks and the financial support from Sim
Kee Boon Institute for Financial Economics in Singapore, as well as the Geneva Finance Research Institute
(GFRI). We also acknowledge a substantial research contribution by Choong Tze Chua at an initial phase of the
project. Finally, we thank seminar participants at the 2012 annual meeting of the Western Finance Association
(WFA), European Central Bank, INSEAD, London School of Economics, University of Cologne, University of
Hong Kong, University of Maryland, University of Rotterdam, University of Virginia, Singapore Management

University, and the 2011 European Summer Symposium in Financial Markets (ESSFM) for their comments on
an earlier draft of the paper. Special thanks go to George O. Aragon (the WFA discussant), Miguel A. Ferreira,
John Grin, Denis Gromb, Gerard Hoberg, Andrew Karolyi, Pete Kyle, Roger Loh, Francis Longsta, Alberto
Manconi, Gordon Philips, Christopher Polk, Kalle Rinne, Mark Seasholes, Mitch Warachka, Francis Warnock,
K.C. John Wei, Kelsey Wei, Eric Wincoop, and Dimitri Vayanos.

The Role of Equity Funds

in the Financial Crisis Propagation

July 25, 2014

Abstract
The early stage of the 200708 financial crisis was marked by large value losses
for bank stocks. This paper identifies the equity funds most aected by this valuation shock and examines its consequences for the nonfinancial stocks owned by
the respective funds. We document three key empirical findings. First, ownership
links to these "distressed equity funds" lead to large underperformance of the most
exposed nonfinancial stocks. The results hold even after controlling for various accounting valuation measures. Second, distressed fund sales and the associated price
discounts are concentrated among those exposed stocks that perform relatively well.
Third, stocks with higher overall fund ownership generally performed much better
throughout the crisis.
JEL Classification: G11, G14, G23
Keywords: Financial Crisis, Crisis Spillover Eect, Mutual Fund Ownership

Introduction

Financial sector stocks accounted for only 20% of the total U.S. stock market value in 2007.
Their widespread exposure to the subprime market not only hurt their own stock prices, but
eventually led to a near 50% value decrease for nonfinancial stocks as well.1 Is the price drop
in nonfinancial stocks fully justified by their fundamentals, or is there a price contagion from
financial to nonfinancial stocks? Using fund ownership data at the stock and fund/investor
level, this paper identifies the common fund owners between financial and nonfinancial stocks
as an important channel for price contagion during the crisis. Our analysis suggests that at
least one fifth of the 53% crisis-related decline for the U.S. stock market is attributable to price
contagion via such common fund ownership. By examining the 200708 crisis development in
the stock market from this new angle of joint equity fund ownership between crisis and noncrisis
stocks, our study identifies a sharp macroeconomic picture of crisis-induced transitory equity
price dynamics. In particular, we highlight that the propagation of financial instability does not
require leverage of financial intermediaries and that leverage regulation alone may not always
be sucient for crisis containment (Shin, 2013; Kashyap, 2014).
A large empirical literature documents price contagion across countries and asset classes.2
Yet, as Forbes and Rigobon (2002) argue, it is often dicult to separate contagion from ordinary
asset interdependence. This paper uses a new comprehensive sample on the equity positions
of 22 621 equity funds around the world for a clear identification of a contagion channel. For
each fund, we calculate fund exposure to financial stocks as the funds return losses induced by
financial sector positions in the initial phase of the financial crisis. Distressed equity funds with
large losses faced larger investor redemptions and, therefore, had to engage in asset fire sales
of their financial and/or nonfinancial stocks. To capture the selling pressure of a nonfinancial
stock owned by distressed funds, we define its stock exposure as the ownership-weighted average
fund exposure of all mutual funds owning the stock. Thus, nonfinancial stocks held by funds
1

As of June 2007, financial stocks (SIC codes between 6000 and 6799) had a total market capitalization
value of about US$3 771 billion, compared to US$13 624 billion for nonfinancial stocks. By February 2009,
their respective values dropped to US$1 010 billion and US$7 176 billion, which represent a value decline of
73% for financial stocks and 47% for nonfinancial stocks. The overall U.S. stock market value decreased by
about 53% during this period.
2
See Kindleberger (1978); Dornbusch, Park, and Claessens (2000); and Kaminsky, Reinhart, and Vegh (2003)
for excellent surveys.

with heavy loadings on underperforming financial stocks would be considered highly exposed
stocks, which have high exposure to the financial sector via their fund owners.
Separation of the stock universe into financial and nonfinancial stocks allows for a better
identification of causal eects. We do not condition our analysis directly on fund outflows
because of concerns about outflow endogeneity in the context of the crisis. Funds with poor
overall performance are likely to experience larger outflows, so that conditioning the analysis on
fund outflows would generate a sample bias toward funds holding underperforming (financial
and nonfinancial) stocks. To avoid such a selection bias, we capture fund distress as (ex post)
poor asset allocations in financial sector stocks only, while measuring fire-sale eects exclusively
for nonfinancial stocks.3
Our empirical analysis focuses on the relative return of exposed stocks, i.e., the 30% of
nonfinancial stocks with the highest stock exposure. We show that nonfinancial stocks with
high exposure to distressed funds underperformed considerably during the financial crisis. The
price for the exposed U.S. stocks underperformed relative to nonexposed industry peers by
409% at the peak of the stock market downturn. This highlights the role of funding constraints
for mutual funds and their importance for stock market contagion.
In our research design, we carefully exclude stocks in banking-related industries (e.g., banking, insurance, real estate, and financial trading) from our sample of nonfinancial stocks. In
addition, we exclude conglomerates that have more than 1% of total sales in these bankingrelated industries. Our findings are also robust to the control of various firm characteristics and
accounting measures, including the Amihud illiquidity measure, receivables-to-sales ratio, priceto-book ratio, leverage ratio, short-term debt-to-asset ratio, and dividend yield. Importantly,
our results cannot be explained by any omitted characteristics that are common between financial stocks and exposed, nonfinancial stockssuch an explanation implies the greatest price
discount among the worst-performing exposed, nonfinancial stocks (due to their shared firm
characteristics with financial stocks). Contrary to the omitted variable hypothesis, we find that
the fire-sale discount is most pronounced for those exposed stocks that performed relatively
well during the crisis.
3

Fund outflows also may be driven by a few investors foresight into the future performance of a fund. In this
case, outflows correlate with future stock underperformance, and the fire sale eect becomes entangled with a
confounding selection eect.

The pronounced concentration of fire-sale discounts for the best-performing stocks also suggests that distressed funds preferred to liquidate stock positions for which selling did not imply
realizing large capital losses. As a paradoxical consequence, large transitory stock underpricing
primarily aicted those stocks that had no real crisis exposure other than being owned by the
distressed equity funds with large exposure to the financial sector. This has further implications
for macroeconomic research on the real eects of a financial crisis. Such research might easily
arrive at biased results for the real transmission mechanism unless it properly controls for fund
ownership linkages. Hau and Lai (2013) provide evidence that justifies such a concern. They
show that stock underpricing induced by fund ownership linkages reduced firm investment and
employment substantially during the 200708 crisis. Unlike the study by Hau and Lai (2013),
which focuses on the real eect of stock underpricing, the current paper provides a comprehensive analysis of the propagation of the crisis across dierent stock classes in the U.S. and
abroad. The subprime crisis started out in the U.S. with an initial impact on bank stocks,
which was then propagated through joint equity fund ownership to other non-financial stocks in
the U.S. and in other countries. The fund ownership channel identified in this paper accounts
for a relative underperformance of exposed stocks of 261% in other developed countries and
175% in emerging countries at the peak of the crisis.
Furthermore, we find that while ownership by distressed funds adversely aected the performance of a stock during the crisis, the opposite is true for nondistressed fund ownership. Stocks
in the top 30% quantile of the highest aggregate overall fund ownership suered considerably
lower capital depreciation than stocks in the bottom quantile. Could this be due to the dierent
flight to quality propensity between retail investors and other types of investors? Kumar and
Lee (2006) show that retail investors typically concentrate their holdings and trading activities
in stocks with low institutional ownership rather than those with high institutional ownership.
If retail investors generally have a higher propensity for flight to quality during a crisis than
institutional or fund investors, this might contribute, at least partially, to the underperformance
of stocks with low overall fund ownership.4
4

Although the majority of equity fund shareholders are also retail investors, Bailey, Kumar, and Ng (2011)
show that those who are willing to delegate their portfolio decisions tend to dier in their socioeconomic and
behavioral characteristics, compared to those who manage their own capital directly (i.e., direct retail investors).
Therefore, it is possible that direct retail investors feature a dierent propensity of "flight to quality" than equity

Using a daily vector autoregression (VAR) system, we find evidence consistent with such an
argument. Specifically, we construct a long-short portfolio (which has long positions in stocks
with the lowest fund ownership and short positions in stocks with the highest fund ownership)
to capture the flight to quality sentiment of retail investors relative to other investors. We find
that (negative) market-wide shocks Granger caused the return of the long-short portfolio during
the crisis period, but not the pre-crisis period. An alternative long-short portfolio constructed
based on the retail buy-and-sell volume of NYSE stocks obtained from NYSE Euronext also
yields qualitatively similar results.
This study adds to a nascent literature that uses portfolio data to identify channels of asset
contagions. In particular, Broner, Gelos, and Reinhart (2006) find that rebalancing toward the
index (retrenchment) by global equity funds during the previous emerging market crises (i.e.,
Thailand in 1997, Russia in 1998, and Brazil in 1999) had a pronounced eect on the crosssection of international equity index returns. Manconi, Massa, and Yasuda (2012) find that
during the 200708 crisis, fixed-income mutual funds transmitted the crisis from the securitized
bond market to the corporate bond market. Badertscher, Burks, and Easton (2012) find little
support that the mandatory bank asset value write-down during the recent financial crisis
results in banks fire-selling debt securities. Our study focuses on another group of institutional
investors (i.e., equity mutual funds) and finds that the initial value losses in some equity funds
lead to significant asset fire sales, which substantially worsen the crisis.
Other works have taken a broader approach to characterize contagion channels. Calomiris,
Love, and Martinez Peria (2012) examine how the collapse of global demand, the contraction
of credit supply, and the selling pressure of firm equity jointly depressed non-U.S. stock prices
in the 200708 crisis. They used a stocks free float and stock turnover as measures of asset
liquidity and proxies for equity selling pressurea weaker identification scheme than the stock
exposure measure we propose in this paper. Longsta (2010) provides complementary evidence
on contagion from the ABX subprime indices to the bond market and financial stocks. Bekaert
et al. (2012) focus on the international transmission of financial crisis and identify crisisrelated risk factor changes. By contrast, the price eects we document are based on ownership
characteristics of individual stocks, instead of relying on the more simplified factor structure
fund managers.

representation. Similar to Bartram, Grin, and Ng (2012), we argue that ownership linkages
are a highly important driver of stock returns, especially during a financial crisis.
Our analysis also relates to a growing body of literature on stock market mispricing and
limits of arbitrage surveyed by Kothari (2001), Lee (2001), and Gromb and Vayanos (2010).
This literature has highlighted the role of funding constraints of financial intermediaries in
determining asset prices. Financial crises may give rise to a greater and more pervasive asset
mispricing. For example, Rinne and Suominen (2010) show that asset liquidity in U.S. stocks
generally dropped during the 2007/08 crisis. A more extensive arbitrage breakdown may arise
endogenously from larger asset valuation complexity if a crisis generates new unknown liquidity
externalities (Caballero and Simsek, 2011). As a result, limits of arbitrage may shift during a
crisis. The large-scale fire sale discounts documented in this paper suggest such a displacement
of arbitrage boundaries.
Section 2 discusses the data. Section 3 describes this papers research design and method.
Section 4.1 presents evidence for the fire-sale discounts along the timeline of the crisis. Section
4.2 uses quantile regressions to document the asymmetric eect of fire-sale discounts by stock
performance quantiles. Section 4.3 examines the return eect of dierential firm characteristics
between exposed and nonexposed stocks. Section 4.4 presents evidence of distressed fund
selling that matches the return evidence. The hypothesis of dierent propensities for flight to
quality between retail investors and other investors is examined in Section 4.5. Evidence on
international propagation of the crisis is presented in Section 4.6. The robustness issues related
to stock selection biases are examined in Section 5. Section 6 concludes.

Data

Our analysis focuses on the crisis-related underpricing of U.S. nonfinancial stocks. except for the
international evidence presented in Section 4.6. Our measure of stock exposure is based on the
worldwide fund holdings data from the Thomson Reuters International Mutual Fund database.
The use of worldwide fund holdings is warranted because foreign funds hold a nonnegligible
share of U.S. stocks. A detailed description of this database is provided in Covrig, Defond,
and Hung (2007). They use the data from 19992002 to examine whether the adoption of

International Accounting Standards are able to attract more foreign capital, whereas we use the
data from 20072009 to examine the transmission of the subprime mortgage crisis from financial
stocks to nonfinancial stocks. The Thomson Reuters data account for both pure equity funds
and the equity holdings of balanced funds, which also hold other assets, such as bonds. In the
latter case, only the equity portion of the fund holdings is reported. Most international funds
only report at six-month intervalshence, the analysis related to fund holdings is carried out
on a semi-annual basis. For funds with multiple reporting dates within a semester, we retain
only the last reporting date.5
Based on fund holdings data, we remove funds that had more than 75% of their asset
holdings in financial stocks because these funds are likely to be financial sector funds. For
those funds, the investment on banking stocks might be nondiscretionary, so investors might
not attribute underperformance to a poor fund allocation. We therefore exclude such funds
from the sample and focus on those with discretionary investment in financial stocks. We also
exclude index funds and ETFs from our sample.6 A general index selling by institutions is not
likely to aect exposed and nonexposed stocks dierently because, presumably, index selling
does not distinguish between these two types of nonfinancial stocks. Our final sample includes
a total of 22 621 funds reporting stock positions with a combined total net equity value of
US$97 trillion as of June 2007.
We obtain the daily, weekly, and monthly global stock return data from Datastream. All
return calculations are based on the total return index to account for dividend payments and
capital measures. Global banking stocks are defined based on Datastream industry code 102.
In order to have a cleaner measure of the crisis transmission eect from financial to nonfinancial
stocks, we remove banking-related industries from the sample of nonfinancial stocks. Based on
5

We can compare the Thomson Reuters aggregate country holding data to the ICI international fund statistics. The correlation between the holdings reported by Thomson Reuters and those reported by ICI (in logs of
million dollars of equity fund assets) is 87.6% across countries. For the U.S. and Canada, the aggregate equity
positions reported by Thomson Reuters dier from ICI by only -0.26% and 0.82%, respectively. To conserve
space, the detailed comparison between the two databases is not tabulated but is available from the authors
upon request.
6
Because there is no index fund indicator in our fund database, we screen the names of all funds. If the
word "index" or "ETF" appears in a funds name, the fund is removed from our sample. We concede that such
a keyword search may not fully purge index funds from our sample, but a general index fund selling is unlikely
to explain our empirical findings.

the Fama and French 48 industry classification, we identify banking, insurance, real estate, and
financial trading as banking-related industries. We use SIC codes to identify U.S. stocks in
these industries. For international stocks, we use Datastream industry codes 36, 42, 46, 77, 85,
102, 106, 108, 111113, 133, 141, 152, 160167, and 184 to identify them. In addition, using
the Compustat industry segment file, we further exclude conglomerates that have more than
1% of total sales in those banking-related industries.
To account for the dierence in firm characteristics among stocks, we obtain the market
capitalization and the price-to-book ratio from Datastream based on the most recent data
available, as of June 2007. The receivables-to-sales ratio, leverage (total debt-to-asset) ratio,
short-term (ST) debt-to-asset ratio, and dividend yield are obtained from the Compustat database, based on the latest fiscal year-end data prior to July 2007. In addition, we calculate the
Amihud illiquidity measure (Illiquidity) as the ratio of the daily absolute stock return to the
dollar trading volume, averaged over July 2006 to June 2007. Panel A of Table 2 shows that
the 30% most exposed U.S. nonfinancial stocks tend to be larger and more liquid than the
rest of U.S. nonfinancial stocks (i.e., nonexposed, nonfinancial stocks). This corresponds to the
general finding that fund ownership is biased toward larger and more liquid stocks; such characteristics should attenuate any return eect fund sales may have on exposed stocks. On average,
the exposed stocks also have higher leverage but lower receivables-to-sales, price-to-book, and
short-term debt-to-asset ratios than nonexposed stocks.

Research Design and Method

3.1

Hypotheses

The first fallout of the subprime crisis in 2007 was a substantial value loss for bank stocks. The
mean return for U.S. financial stocks in the second semester of 2007 and the first semester of
2008 was a catastrophic 274% and 325% respectively.7 As a consequence, equity funds
with large shares of ownership in financial stocks suered a substantial negative shock to their
7

See Gorton (2008) for a detailed discussion of the crisis chronology. An important public signal at the
beginning of the crisis was the downgrading of mortgage-backed securities by S&P and Moodys on July 10,
2007. The returns of 274% and 325% for U.S. financial stocks are calculated based on the S&P1500 Banking
Index from June 29, 2007 to December 28, 2007 and June 27, 2008.

fund performance. These are likely to face stronger fund outflows after large value losses.
The so-called fund flow-performance relationship has been extensively documented in the
literature (see, for example, Chevalier and Ellison (1997) and Sirri and Tufano (1998)). To
meet redemption requirements from investors, such equity funds might have to liquidate part
of their portfolio, which in turn depreciates the equity values of the stocks they sell.8 The
distressed selling of nonfinancial stocks by their fund owners, therefore, eectively spreads the
crisis from financial sector stocks to nonfinancial sector stocks.
We first explore whether the common fund ownership between financial stocks and nonfinancial stocks during the 2007 financial crisis represents a channel of price contagion from the
former to the latter. We call this hypothesis the Simple Fire Sale Hypothesis. If this hypothesis
holds, we expect that the subset of nonfinancial stocks linked by common fund ownership to
poorly performing financial stocks would underperform during the financial crisis. Furthermore,
such price contagion should lead only to temporary mispricing, which we expect to fully revert
in the long run.
Empirically, we can test this hypothesis by defining a stock exposure dummy, which marks
those nonfinancial stocks that have distressed equity funds as the principal owners. Fund
distress itself can be measured by the return loss experienced by a fund due to investments
in financial stocks in the initial stage (the second semester of 2007 and the first semester of
2008) of the crisis. The simple fire sale hypothesis also predicts that given the initial holdings
position at the onset of the crisis, the aggregate fund holdings should decrease more strongly
for exposed, nonfinancial stocks than for nonexposed, nonfinancial stocks. We will subject this
hypothesis to rigorous tests in Section 4.
The above hypothesis does not discriminate between the types of stocks a distressed equity
fund may choose to sell. There are reasons suggesting that funds may choose to first sell betterperforming stocks than poorly performing stocks. First, if stock prices generally feature more
pronounced deviations from fundamental values during a crisis, then a simple heuristic decision
rule suggests that a fund first sells stocks with the highest-realized crisis returns because other
(relatively poor-performing) stocks provide hope for a later price reversal. This implies that
8

See also Pulvino (1998) and Coval and Staord (2007) for related evidence that fire sales by distressed firms
or equity funds produce lower asset values.

stocks in the higher performance quantiles are more likely to suer from temporary underpricing. Second, U.S. tax law encourages mutual funds to pass on capital gains from asset sales to
investors because the marginal tax rate for funds is typically higher than the rate for investors.
To further minimize investors capital gains taxes, fund managers may have an incentive to
realize capital gains during the market downturn, when fund investors might have more capital
losses from elsewhere to oset these gains. (Seida and Wempe, 2000.)9
By contrast, the fund window dressing literature (see, for example, ONeal (2001), Meier
and Schaumburg (2006), and Sias (2006)) argues that poorly performing funds are particularly
prone to concealing their poor stock picks by replacing underperforming stocks with overperforming stocks just before they report their asset holdings, suggesting that stocks in the lower
performance quantiles are more likely to be sold by distressed funds. Ultimately, whether
funds condition their equity sales on the crisis performance of a stock is an empirical question.
Therefore, we also examine the Stock Performance-Dependent Fire Sale Hypothesis in Section
4.
A straightforward procedure to explore this hypothesis is to measure the fire sale eect
for dierent stock performance quantiles. We can also directly compare the decrease in fund
holdings for the exposed stocks that performed relatively well to those that performed relatively
poorly during the crisis. Stock liquidity consideration may also matter for the choice of stocks
for fire sales, but a fire-sale preference for more liquid stocks needs not translate into a fire sale
discount because those stocks are generally more price-resilient. Nevertheless, we control for
stock liquidity in our empirical tests.
Our last hypothesis concerns the return eect of overall equity fund ownership. While
distressed funds may have a negative influence on the crisis performance of the stocks that
they initially own, such a negative eect seems unlikely to pertain to equity fund ownership in
general. In fact, the opposite may hold. As shown by Kumar and Lee (2006), retail investors
tend to concentrate their holdings and trading in stocks with low institutional ownership, as
opposed to stocks with high institutional ownership. If retail investors generally exhibit a higher
propensity for panic sales or flight to quality than institutional investors or fund investors
9

If equity fund managers suer from a behavioral bias commonly referred to as the disposition eect, they
will also be more likely to liquidate better-performing stocks than underperforming stocks. Frazzini (2006)
shows that such a behavioral bias exists among equity funds, particularly distressed funds.

during a crisis, then stocks with high retail ownership or low fund ownership can underperform
other stocks. In this case, nondistressed equity fund ownership serves as a stabilizing force in
the stock market during a crisis.
We examine this Fund Share Stability Hypothesis in a cross-sectional regression analysis
of crisis returns on a stocks overall fund ownership. In addition, we construct a long-short
portfolio (which has long positions in the 30% of stocks with the lowest fund ownership and
short positions in the 30% of stocks with the highest fund ownership) to capture the flight to
quality sentiment of retail investors relative to other investors. Using a daily VAR system,
we examine the role of lagged return shocks of the U.S. stock market index on the relative
performance of stocks with low vs. high overall fund ownership. The Fund Share Stability
Hypothesis suggests a strong, positive cumulative impulse response of the long-short portfolio
to (negative) index return shocks during the crisis.

3.2

Fund Exposure and Stock Exposure

We measure stock exposure in two steps. In the first step, we identify a funds exposure to
financial stocks. Global banking stocks are defined based on Datastream industry code 102.
Without loss of clarity, we use the term financial stocks and banking stocks interchangeably
in the following discussion. Let () denote the number of shares held by fund in stock
at time , and () denote the corresponding stock price. The portfolio share of fund (for
the equity components of its investments) in stock is as follows:
() ()
X
() =

() ()

We calculate the financial stock-related return of fund as its value loss over a semester
attributable to financial stock positions; hence:
=

1
() + ( 1) ,
2

where denotes the semester stock return, and the summation involves all financial sector
stocks worldwide. The average return is measured for the arithmetic midpoint between the
beginning and the end of semester weights. Fund exposure for is defined as its return loss due
10

to financial stock investments, and funds without any return loss are deemed to have a zero
fund exposure. That is,
() = min( 0)
Highly negative fund exposure can result from large portfolio weights for bank stocks in general
and/or portfolio holdings in banks with particularly low returns. The identification of the
valuation shock focuses on two semesters from July 2007 to June 2008, before the subprime
crisis became a general financial crisis with the collapse of Lehman Brothers on September 15,
2008. The fund exposure for the second semester of 2007 is denoted by (20072) and for
the first semester of 2008 by (20081) The total fund exposure, , is measured by the
combined return losses over the two semesters:
= (20072) + (20081)
The mean (median) fund exposure is 264% (217%) with a skewness of 23. The 25%,
15%, and 10% lowest fund exposure quantiles are given by 394% 497% and 580%
respectively.
In the second step, for each nonfinancial stock , we aggregate the exposure of their
fund owners to an ownership-weighted measure of stock exposure. Let

( ) = X

denote the ownership share of fund relative to the aggregate ownership of all funds in stock
in June 2007, and denote the aggregate ownership of all funds in stock relative to
the stocks market capitalization in June 2007. The exposure of a nonfinancial stock to the
financial sector (via equity fund ownership) can then be defined as:
=

()

A high stock exposure ( ) implies that a relatively large proportion of a stocks capitalization
is owned by equity funds with high exposure to financial stocks. Such high-exposure stocks

11

should, therefore, face the largest selling pressure if fund exposure captures the need for fire
sales by individual funds.
Summary statistics on U.S. stock exposure are reported in Table 1. The mean (median)
stock exposure is 027% (018%) with a skewness of 15. The 25%, 15%, and 10% most
negative stock exposure quantiles are 043%, 056%, and 065%, respectively. For example,
a stock exposure of 043% will be obtained if 10% of a stocks capitalization is owned by funds
that, on average, lost 43% of their portfolio returns from financial stock investments.
The distribution of stock exposure is highly skewed and its eect on return and holding
change might be nonlinear. It is therefore useful to define a dummy variable that
marks all stock exposures below a certain quantile ( ) where

1 for
( )

=
0 otherwise

Our empirical analysis focuses on the 30% quantile.10 We define the 30% U.S. stocks with most
negative as exposed stocks, and the remaining 70% as nonexposed stocks. For expositional
purposes, we can also define exposed funds and nonexposed funds analogously, but based on
. Panel B of Table 2 shows considerable dispersion in the number of funds investing in the
two types of stocks. For the subsample of exposed (nonexposed) stocks, the average number
of exposed and nonexposed fund owners are, respectively, 109 and 131 (15 and 37) funds in
June 2007. Such coarse fund ownership across stocks translates into a large dispersion of stock
exposure.

3.3

Fund Holding Change and Aggregate Holding Change

The fund ownership data allow us to directly observe holding changes. Let () denote the set
of funds with positive holdings in stock in June 2007. The percentage fund holding change
of () in stock over semesters (from to + ) can be expressed as
() =

( + ) ()
100
()

10

Using a continuous stock exposure variable in place of the exposure dummy also gives qualitatively similar
results.

12

The aggregate ownership-weighted average (percentage) fund holding change for stock , over
semester, can then be calculated as
P
P
( + )
()
X

()

()
P
100 =
( ) ()
() =
()
()

()

We can further define the stock capitalization scaled aggregate (percentage) holding change as
e () = () =

( ) ( )

()

where the product () denotes the ownership share of fund in stock relative to
the total capitalization of the stock.
The aggregate fund holdings decrease over consecutive semesters for U.S. nonfinancial stocks
is shown in Table 1. The average aggregate holding change for = 1 2 3 4 5 is given by 26%
48% 64% 74% and 88% respectively. Section 43 explores whether this aggregate
fund holding decrease is more pronounced for stocks with distressed fund owners.

3.4

Risk Adjustment of Returns

Our analysis of the fire sale eects on stock prices first removes risk premia from the return
analysis. For this risk adjustment, we use the international version of the Carhart (1997)
four-factor model. For each country, we construct a domestic and an international version of
the four factors: The market factor ( ), the size factor (), the book-to-market factor
(), and the momentum factor (). The factor construction is based on monthly stock
returns in U.S. dollars over the five-year period from July 2002 to June 2007. We calculate
the international factors of a country as the weighted-average domestic factors of all other
countries. The weights are given by the relative stock market capitalization of each country at
the beginning of the year.
We estimate the loadings of each stock on the domestic and international risk factors
( = ) using a regression over 60 months, from July 2002 to June 2007. With the
estimated factor loadings, we can calculate the expected return for each stock in month
during the crisis period, July 2007December 2009. The cumulative expected return over
weeks (since month ) is then calculated as follows:
13

 () = (1 + +1 )4

(1 + + ) 1

=1

where denotes the number of full months (since month ) and the number of weeks falling
into the last month + 1 The cumulative risk-adjusted excess return of stock over weeks
can be calculated from the weekly stock return () and the estimated expected return as
() =

(1 + + ) (1 + ())

=1

The cumulative risk-adjusted excess return of stock over semesters (or 6 months) can
be calculated in a similar manner as
() =

6
Q
=1

(1 + + )

6
Q

(1 + + )

=1

The summary statistics for cumulative risk-adjusted returns of all U.S. nonfinancial stocks
are stated in Table 1. The standard deviation of cumulative excess returns increases from 0607
to 1483 as the return horizon under consideration increases from one semester (December
2007) to three semesters (December 2008). The cumulative excess return dispersion decreases
thereafter to 1112 and 1065, respectively, as we consider returns extending until June 2009
and December 2009. This reveals some degree of excess return reversal for nonfinancial stocks
in 2009. We describe the factor construction and the estimation of the expected returns in
detail in the appendix.

4
4.1

Evidence
Stock Exposure Eects on the Crisis Timeline

Did losses in financial stock investment by a fund aect the performance of other stocks (or
nonfinancial stocks) held by the same fund? A simple OLS regression of the risk-adjusted
returns () over semesters of all nonfinancial stocks on the dummy variable can
reveal the role of distressed fund owners in the crisis performance of a stock:
() = 0 + 1 + 2 +

14

The dummy variable denotes the 30% U.S. nonfinancial stocks with the highest distressed fund ownership. Similarly, we define a dummy for the 30% U.S. nonfinancial
stocks with the highest overall fund ownership relative to the stock capitalization in June 2007.
If common fund owners facilitate the transmission of crisis from financial stocks to nonfinancial
stocks, we should observe 1 0. serves as a control variable because higher overall
fund ownership allows for more exposure to distressed funds. The regression discards the 1%
highest and lowest return outliers. We include industry-fixed eects in the regression. The
coecient 1 therefore captures risk-adjusted fire-sale discounts over semesters for the 30%
most exposed nonfinancial stocks relative to other nonfinancial stocks in the same industry.
Table 3 reports the regression results for U.S. stocks. Column 1 is for the return period
from July 1, 2007 to December 31, 2007, in which the stock exposure dummy (20072)
is based on contemporaneous fund return losses in the second semester of 2007. The exposure
dummy reveals an underperformance of 116% after one semester in December 2007, 153%
after two semesters in June 2008, and 206% after three semesters in December 2008. For
June 2009 (after four semesters) we find a reversal of the discount to 96%, and by December
2009 (after five semesters) the discount is no longer significantly dierent from zero. The high
fund-ownership dummy shows a significantly positive coecient. Therefore, stocks with
high overall fund ownership experience better crisis performancea finding consistent with the
Fund Share Stability Hypothesis. We explore this hypothesis further using a daily VAR system
in Section 4.5.
Figure 1, Panel A, plots the coecient for the exposure dummy and a confidence
interval (of 1 standard deviation) using cumulative risk-adjusted returns with weekly return
increments. The regressions after 26, 52, 78, 104, 156 weeks coincide with regressions after
= 1, 2, 3, 4, 5 semesters. The corresponding dates for the five end-of-semester regressions
are highlighted by dashed vertical lines. The fire sale eect shows negative twin peaks around
November 7, 2008, and February 27, 2009, with an average return shortfall of 315% and
409%, respectively, for exposed stocks. The estimation results for the twin peaks are reported
in the last two columns of Table 3.
Our results highlight that crisis propagation through fund exposure played a quantitatively
important role for the overall index decline during 2007/09. An incremental return shortfall of

15

409% for the 30% exposed stocks implies an aggregate eect of 10% (= 409% 30% 80%)

value decline for an equally weighted U.S. stock index.11 Considering the fact that exposed

stocks are, on average, larger than nonexposed stocks, the contribution of this eect to the
decline of the overall U.S. stock market index, which is value-weighted, is likely to be at least
as large. It is therefore not surprising that the maximum fire sale eects identified above are
close to the two weekly U.S. stock index minima on November 7, 2008 and March 6, 2009.

4.2

Stock Exposure Eects by Stock Performance Quantile

Discretionary liquidation of stock positions by distressed funds implies a Stock PerformanceDependent Fire Sale Hypothesis. We therefore estimate regressions for the 25%, 50%, 75%,
and 90% quantiles of the cumulative excess return distribution as a linear function of the stock
exposure dummy . We use November 7, 2008 and February 27, 2009 as the reference
dates for the cumulative returns because they represent the twin peaks of the fire-sale discounts
as shown in Figure 1. As before, the regressions controls for the fund ownership dummy
and industry-fixed eects. In addition, we control for stock liquidity proxied by either
(a dummy for the 30% most liquid U.S. stocks based on the Amihud illiquidity measure) or
(the natural logarithm of firm size).
Table 4 reports the regression results. When controlling for both liquidity proxies in February 2009, the coecient of the stock exposure dummy decreases from 38% and 77% for the
25% and 50% quantiles to 325% and 808% for the 75% and 90% quantiles, respectively.
A similar pattern is observed for the earlier crisis peak in November 2008. Therefore, the
stock exposure measure has an extremely asymmetric eect on the distribution of cumulative
stock returns, with the most negative impact found for the best-performing stocks. The result
suggests that when faced with funding constraints and investor redemption requirements, distressed equity funds first liquidated the best-performing stocks, rather than stocks with recent
large capital losses.
11

U.S. nonfinancial stocks accounted for around 80% of the U.S. stock market in June 2007.

16

4.3

Stock Exposure Eects and Firm Characteristics

Could the stock exposure eects we document in the previous subsections be due to the differences in firm characteristics between exposed and nonexposed stocks? We explore such a
possibility in this section. Firms with a higher receivables-to-sales ratio can be more adversely
aected by the liquidity crunch experienced by the commercial paper market during the crisis.
A high price-to-book, leverage, or short-term debt-to-asset ratio can indicate the vulnerability
of a firm during the crisis, due to a higher default risk. Firms with a higher dividend yield may
experience a higher before-tax stock return. A higher stock illiquidity can amplify the fund
sale eect. We therefore include these stock characteristics as additional controls. The results,
reported in Table 5, indicate that the receivables-to-sales ratio, price-to-book ratio, leverage,
and dividend yield show no reliable evidence of explanatory power for the cross-section of cumulative stock returns in our sample. Not surprisingly, higher stock illiquidity and short-term
debt-to-asset ratios are associated with more negative crisis returns. However, controlling for
these firm characteristics has no qualitative eect on the results reported in Table 3. Therefore, we conclude that dierences in firm characteristics between exposed and nonexposed stocks
cannot account for the fire sale eect measured by the stock exposure dummy.12
Could any omitted firm characteristics that are common between financial stocks and exposed, nonfinancial stocks explain our findings? As discussed in the data section, we try to
eliminate such a possibility by removing banking-related industries from our sample of nonfinancial stocks. We also exclude conglomerates that have more than 1% of total sales in those
banking-related industries. Importantly, any common omitted firm characteristics between
financial and nonfinancial stocks would imply the greatest price discounts among the worstperforming nonfinancial stocks (due to their shared firm characteristics with financial stocks).
However, contrary to the omitted variable hypothesis, the evidence presented in Section 4.2
shows that the fire sale discount is most pronounced for those exposed stocks that performed
relatively well during the crisis.
12

We also test whether time-changing risk premia and factor loadings can explain our findings. Specifically,
we include stock betas as additional control variables in the cumulative return regressions of Table 3. We find
that such an extended specification does not qualitatively alter our regression results.

17

4.4

Fund Holding Changes

When facing liquidity constraints or anticipating investor redemption, exposed equity funds
e () the (percentage) aggregate holding
were likely to engage in fire sales. We denote by

change in stock over semesters of all funds with initial positions in June 2007. Analogous to
the return regression, the holding change is related to the dummy variables and .
The 1% of smallest and largest holding changes are discarded from the linear regression given
by
e () = 0 + 1 + 2 +

The fire sale hypothesis implies 1 0 because exposed stocks should show a faster holding
decline for the initial owners in June 2007. To test for the stock performance-dependent fire
sale hypothesis, we extend the above specification by a dummy variable , marking
all U.S. stocks in the 30% quantile with the highest cumulative return over the semesters
since June 2007. A second dummy is defined as the product of the stock

exposure dummy and the high-return dummy The extended specification

becomes
e () = 0 + 1 + 2 + 3 + 4 ( ) +

where the interaction term captures incrementally larger holding reductions for those exposed
stocks that do relatively well during the crisis. More pronounced position liquidations in these
stocks imply 4 0
Table 6 provides the regression results for U.S. stocks. For each incremental semester,
we first report the baseline specification and then the extended specification. Exposed stocks
(with = 1) show an accelerated decrease in the aggregate holdings by funds that are
initial owners in June 2007. The additional cumulative decrease amounts to 085%, 161%
216% and 270% over a period of = 1, 2, 3, 4 semesters, respectively. Compared to
the average holding decreases of 256% 478% 640% and 740% (reported in Table 1),
these figures reveal approximately 35% more net fund-selling for the 30% most exposed stocks
than for an average stock.

18

The dummy interaction term is statistically significant and shows that

exposed stocks with good crisis performance had more dramatic holding reductions. The incremental holding decrease captured by the coecient 4 is 034% 092% 123% and
133%, relative to 084% 145% 177% and 216% measured by the coecient 1

The ratio of 133% to 216% suggests a 60% greater decrease of exposed stock holdings if
the stock was among the 30% best-performing stocks. This finding matches the return evidence
from the quantile regressions in Table 4 and supports the stock performance-dependent fire sale
hypothesis.
e aggregates holding changes only for those funds that initially
We note that the variable

hold a strictly positive position in the stock in July 2007. Therefore, the coecient for

is strongly influenced by a general mean reversion property of the high fund share variable,
suggesting a persistently negative sign for the coecient of . Indeed, we find the
estimate to be significantly negative. An alternative explanation comes from the finding in the
literature suggesting that stocks with low fund ownership typically feature more retail ownership
and trading (Kumar and Lee (2006)). Suppose retail investors engage in panic sales during the
crisis and equity funds act as the liquidity supplier. In that case, we are likely to observe more
funds buying stocks with low fund ownership (or high retail ownership), again suggesting a
negative sign for . We explore this hypothesis further in the next subsection.
e (4), from July 2007 to
Figure 2 compares the distribution of cumulative holding changes

June 2009, between exposed and nonexposed U.S. nonfinancial stocks. Exposed stocks feature

a much larger left-tail distribution, indicating that large aggregate holding reductions were
much more frequent for these stocks. Such drastic holding reductions by distressed funds are
consistent with the fund redemption evidence we obtained from the Lipper Fund Database and
matches the return evidence provided in Section 4.1.13
13

We are not able to gauge the exact redemption pressure faced by our sample funds because the Thomson
Reuters International Fund database does not provide any fund flow data. However, for a subset of 8 250 funds,
for which we are able to obtain the fund return and fund net asset value (TNA) data from Lipper, we find that
exposed funds started to experience net investor outflows (estimated using fund returns and fund TNA) after
July 2007. This accumulated to a sizeable average fund outflow of about 7% in late 2008 and early 2009. By
contrast, the average net cumulative inflow for nonexposed funds remained positive from July 2007 to December
2009 .

19

4.5

Retail Panic during the Crisis?

The relative crisis resilience of stocks with high overall fund ownership documented in Section
4.1 is interesting, and it calls for further analysis. Investors typically show a tendency toward
flight to quality in times of market turbulence. Could the flight to quality propensity dier
across investor groups during a crisis? In particular, could retail investors exhibit a higher
propensity for flight to quality than institutional investors or fund investors? Kumar and
Lee (2006) argue that retail investors tend to concentrate their holdings and trading activities in
stocks with low institutional ownership, rather than stocks with high institutional ownership. If
dierent types of stocks feature dierent types of investors, then the dierent flight to quality
propensity of stock investors should translate into dierent return sensitivity of these stocks to
a financial crisis.14
To explore this hypothesis, we construct two long-short portfolios. First, we construct a
(value-weighted) long-short portfolio with long positions in the 30% of stocks with the lowest
fund ownership and a short position in the 30% of stocks with the highest fund ownership using
the 4 663 U.S. stocks in our sample. We call this the (Direct Minus Fund ownership)
portfolio for short. Analogously, we construct a long-short portfolio using the daily retail buyand-sell volume data of NYSE common stocks provided to us by NYSE Euronext.15 After
removing banking-related industries, as well as conglomerates with more than 1% of sales in
these banking-related industries, we are left with 1 036 NYSE stocks that are also in our U.S.
stock sample. We calculate the percentage of retail trading as the sum of monthly retail buyand-sell volume relative to the total trading volume for the stock, averaged over a three-year
period from July 2004 to June 2007. We then construct a long-short portfolio, called the
(retail minus institutional trading) portfolio, which has long positions in the 30% of NYSE
stocks with the highest retail trading and short positions in the 30% stocks with the lowest
retail trading.
Important to the flight to quality phenomenon is a strong reaction to negative past return
14

Studies by Chan, Frankel, and Kothari (2004) and Daniel (2004) also suggest that individual investors are
prone to behavior biases.
15
We thank NYSE Technologies Global Market Data for providing the data.
See
http://www.nyxdata.com/Data-Products/ReTrac-EOD.

20

shocks on the whole market or market index. The long-short portfolio return, or ,
captures the flight to quality sentiment of retail investors relative to other investors. The
daily return on such a long-short portfolio is combined with the daily return on the U.S. MSCI
market index to build a simple structural VAR in = ( ) with innovations
= (1 2 ) of the form
( 1 2 2 + + ) =
where is the lag operator (for lags of ), is the 22 identity matrix, and 1 2 are
unconstrained 2 2 matrices capturing the delayed influence of the lagged dependent variables.
Identification is achieved under the restrictions

1 12
and
=
0 1

11

22

This identification structure allows for a contemporaneous eect of the long-short portfolio
return on the index return because stocks in the long-short portfolio are, by

definition, part of the market portfolio. By contrast, the index return influences the
long-short portfolio only with a lag of one or more trading days. Such a delayed reaction is
particularly plausible for retail investors, who may observe index changes at the end of the day
and only execute their trades on the following trading day. Of particular interest is the role of
index innovations 1 on the portfolio return If retail investors exhibit a higher flight to
quality propensity than other investors, we should expect the stock market index innovations
to have a strong positive eect on
We estimate the VAR for three dierent time periods of 12 months each: 01/07/2006
30/06/2007, 01/07/200730/06/2008, and 01/07/200830/06/2009, which are referred to as
the pre-crisis period, crisis period I, and crisis period II, respectively. The pre-crisis period
serves as a benchmark against which to assess the change in the dynamics between index
returns and long-short portfolio returns during the crisis. For both crisis periods,
the AIC and HQIC criteria indicate that a lag order length = 2 is sucient to capture the
system dynamics. Table 7, Panel A, summarizes the estimation results. The most statistically
significant VAR coecient is (1) in the equation for , which captures the eect of
the index return on at the one-day lag. The parameter estimates are 0294 and 0158
21

(with corresponding z -statistics of 743 and 659) for crisis periods I and II, respectively. Hence,
the portfolio return reacts strongly and positively to the index return on the previous trading
day during the crisis. The causality test shows that the index return is not Granger caused by
the long-short portfolio return. Conversely, there is strong evidence that the stock index return
predicts future returns of the long-short portfolio during both crisis periods, but not during the
pre-crisis period; the respective Wald test statistics for the exclusion of the index return from
the return dynamics of the long-short portfolio are 2 (2) = 362, 2 (2) = 6831, and
2 (2) = 6561 for the pre-crisis period, crisis period I, and crisis period II, respectively.
Figure 3, Panel A plots the cumulative impulse response of the portfolio return to
a unit shock to the index return for the three separate time periods. The pre-crisis period
does not provide any evidence for a stable relationship between index shocks and returns of
the long-short portfolio, as indicated by the wide (95%) confidence intervals. This relationship
changes during the two crisis periods. A 1% innovation to the index return now implies an
average of 042% cumulative return impact on the long-short portfolio during crisis period I,
and still 025% during crisis period II.
Table 7, Panel B reports the corresponding estimation results for the portfolio. The
VAR coecient (1) in the equation for is again economically large at 0119 and
0113 (with z -statistics of 440 and 303) for the crisis periods I and II, respectively. Figure 3,
Panel B plots the corresponding impulse response function of the portfolio following a
unit market return shock. Stocks dominated by retail investor trading show a strong additional
return eect on the days after index shocks during the crisis periods. The cumulative return
eect after five days (to a unit index return shock) is 016 and 009 during crisis periods I and
II, respectively.
In summary, the considerable economic magnitude of the estimated VAR eects suggests
that flight to quality as a reaction to (negative) market-wide shocks concerned directly invested retail equity capital much more than capital managed by institutional investors or delegated portfolio managers. The VAR evidence, therefore, is consistent with the fund stability
hypothesis, which argues that retail investors panic sales during the crisis contribute, at least
partially, to the underperformance of stocks with low fund ownership documented in Table 3.

22

4.6

International Propagation

International stock ownership allows for better global asset diversification but may also create
channels for crisis propagation beyond the U.S. borders. So far, our analysis has examined
ownership-related underpricing only for U.S. nonfinancial stocks, yet it is interesting to extend
the analysis to nonfinancial stocks outside the U.S.
The larger role of mutual funds in the U.S. stock market suggests that stock exposure
through distressed funds is likely to be more widespread and pronounced in the U.S. than in
other countries. Figure 4 plots the stock exposure distribution separately for the 4 663 U.S.
stocks (Panel A), 11 646 developed market stocks ex U.S. stocks (Panel B), and 5 407 emerging
market stocks (Panel C). The international sample spreads across 22 developed markets and
18 emerging markets. As expected, the tail of the stock exposure distribution is fatter for U.S.
stocks, compared to other developed market or emerging market stocks. Nevertheless, both
developed and emerging markets feature a sizable left tail of exposed stocks, for which we can
again define a dummy ( ) marking the 30% most exposed stocks for each country. The
cross-country average stock exposure among these 30% most exposed nonfinancial stocks is
046% and 030% for, respectively, developed market stocks ex U.S. and emerging market
stocks, compared to 063% in the corresponding U.S. stock sample.
Table 8 reports the cumulative risk-adjusted return evidence for all non-U.S. stocks (Panel
A), developed market stocks ex U.S. (Panel B), and emerging market stocks (Panel C) from
July 2007 to the end of each subsequent semester, as well as to the twin peaks of the crisis.
Around the first peak (Nov. 7, 2008), the additional underpricing for the 30% most exposed
non-U.S. stocks amounts to 168%, compared to 315% for U.S. stocks (reported in Table 3).
The corresponding relative underpricing for emerging market stocks is 157%. The weekly
cumulative risk-adjusted returns for international stocks (presented in Panels B, C, and D of
Figure 1) again show a very similar pattern to those for U.S. stocks (in Panel A). Therefore,
international fund ownership linkages played an economically significant role in the international
transmission of the U.S. mortgage market crisis. We conjecture that the gradually decreasing
equity home bias and the globalization of the equity fund industry are likely to make this
international transmission mechanism even more potent in the future.

23

Robustness

Our research design assumes that the ownership concentration of distressed (or exposed) funds
in any nonfinancial stock corresponds to a random treatment eect. The underlying assumption is that the nonfinancial stock picks are not systematically dierent between exposed and
nonexposed funds in terms of the expected stock returns. Hence, concentrated ownership of
exposed funds in any nonfinancial stock becomes a quasi-random coincidence, which does not
feature any performance bias other than the fire sale eect.
To verify this assumption, we first examine whether the exposed, nonfinancial stocks feature
any abnormal returns prior to the crisis, relative to the nonexposed, nonfinancial stocks. Such
abnormal returns can indicate omitted risk factors. Following Fama and French (2010), we form
an equal-weighted portfolio and a value-weighted portfolio separately for the two types of U.S.
stocks each month, from January 2002 to December 2006. We then test for their dierences in
risk-adjusted returns, allowing the risk factor loadings to dier across the two types of stocks.
Table 9, Panel A shows that the abnormal return dierences between exposed and nonexposed
stocks are insignificant after controlling for the standard risk factors in the literature (i.e., the
market, size, book-to-market, and momentum factors). Using a similar approach, we also form
separate value- and equal-weighted portfolios for the nonfinancial stock holdings of exposed
funds and nonexposed funds.16 Panel B of the table again shows that the risk-adjusted return
dierences between the nonfinancial holdings of the two types of funds are insignificant. The
overall result suggests that exposed stocks were not priced according to any omitted risk factor.
We can also examine the similarity of stock portfolios held by dierent types of funds.
Formally, for any pair of funds (1 2 ) we define their portfolio overlap (in nonfinancial sector
stocks) as the minimum common portfolio weight in any stock , summed across all nonfinancial
sector stocks that both funds share; that is
(1 2 ) =

min[
b1
b2 ]

where
b1 and
b2 represent the portfolio weight of nonfinancial stock in funds 1 and 2
16

Specifically, we use a funds nonfinancial stock holdings at the beginning of the semester to estimate its
monthly returns in the subsequent six months. The returns of the equal- and asset value-weighted portfolios
of exposed and nonexposed funds are then calculated separately each month, from January 2002 to December
2006.

24

respectively. We examine such portfolio overlap for the 10% most distressed funds (i.e., funds
with the greatest investment loss from financial stocks from July 2007 to June 2008). We label
this group of funds as Group A. We then match this group with the same number of other
funds, based on their country codes and the size of their total asset holdings in nonfinancial
stocks. The group of matched funds is labeled as Group B. The portfolio overlap statistic is
then calculated for (i) pairs of funds in Group A, (ii) pair of funds in Group B, and (iii) pairs
of one fund from Group A and one from Group B, based on fund holdings in December 2006.
Figure 5 plots the distributions of the three portfolio overlap measures, sorted by percentiles.
All three overlap measures show considerable independence of stock picks across funds. The
average overlap for funds in Group A is 29%, compared to 25% for funds in Group B and
24% for the cross-group pairs. The similarity in stock selections appears to be economically
small for all three groups. In particular, any two funds dier, on average, in 97% of stock
picks, suggesting a limited scope of clustering on stocks with particular unobserved risk factors.
Furthermore, the evidence on the full, long-run price reversal of exposed, nonfinancial stocks
that we present in Figure 1 and Table 3 is another piece of evidence that the distressed equity
funds on average did not pick a below average portfolio of nonfinancial stocks. Therefore, the
underperformance of exposed nonfinancial stocks during the crisis cannot be explained by the
poor nonfinancial stock pick of their distressed fund owners.

Conclusions

Open-end mutual funds have increased their share of the U.S. market capitalization from 46%
in 1980 to 215% in 2007 (French, 2008, p.1539) and have thus become key institutions in equity
markets. Our evidence shows that they played an important role in the transmission of the
2007/08 crisis from financial stocks to nonfinancial stocks, resulting in very large temporary
price discounts for many nonfinancial stocks. This evidence highlights that even non-leveraged
financial intermediaries can play an important role in the propagation of financial instability.
Our identification scheme is based on equity funds return shortfalls induced by financial
sector positions between July 2007 and June 2008. This initial phase of the financial crisis
is marked by dramatic value losses of many bank stocks and the corresponding underperfor-

25

mance of the mutual funds that invest in them. We then study the price externality of such
investment losses in financial sector stocks for the pricing of nonfinancial stocks. For each
nonfinancial stock, we aggregate its fund owners return losses from financial stock investment.
This aggregation results in a measure that captures the financial distress of the nonfinancial
stocks fund owners and, therefore, the selling pressure faced by the stock.
The analysis carefully controls for real linkages between the banking sector and various
industries by excluding banking-related industries, as well as conglomerates that have more
than 1% of total sales in these banking-related industries, from our sample of nonfinancial
stocks. Our findings are also robust to the control of various firm characteristics, including
the Amihud illiquidity measure, receivables-to-sales ratio, price-to-book ratio, leverage ratio,
short-term debt-to-asset ratio, and dividend yield, as well as industry-fixed eects. An analysis
of the 30% most exposed U.S. nonfinancial stocks reveals their dramatic risk-adjusted underperformance. Their relative stock underpricing peaked at 409% in late February 2009, which is
strong evidence that distressed funds played an important role in deepening the stock market
downturn.
Our findings cannot be explained by any common omitted firm characteristics between
financial and nonfinancial stocks because such an explanation would imply the greatest price
discounts among the worst-performing nonfinancial stocks. However, we find that the selling
pressure is greatest for stocks that performed relatively well during the crisis, suggesting that
funds seek to avoid large loss realization from selling the most depressed stocks. We also
find that fund ownership played an important role in the international transmission of the
stock market downturn, even though magnitudes here are smaller because of weaker ownership
exposure links. Specifically, exposed nonfinancial stocks underperform nonexposed industry
peers by about 26% and 18% at the crisis peak for other developed markets ex U.S. and
emerging markets, respectively.
Furthermore, we find that while ownership by distressed funds adversely aected the performance of a stock during the crisis, the opposite holds for overall fund ownership. We explore
whether retail investors generally have a higher propensity for flight to quality during the
crisis than other investors and whether this might have contributed to the underperformance
of stocks with low overall fund ownership (or high retail ownership). Our findings from the

26

impulse response analysis in a daily VAR system are consistent with such an argument.
Overall, we conclude that the fund ownership structure at the outset of the crisis in June
2007 had an astonishingly large eect on the crisis performance of nonfinancial stocks both in
the U.S. and abroad. While better regulation of leveraged financial intermediaries is a necessary
condition for financial stability, it might not be a sucient condition to prevent the propagation
of financial instability.

27

Appendix
This appendix describes the construction of the risk factors. They are based on monthly
stock returns in U.S. dollars from Datastream over the five-year period from July 2002 to June
2007. We exclude non-common stocks such as REITs, closed-end funds, warrants, etc. We
also exclude those firms that are incorporated outside their home countries, as well as those
indicated by Datastream as duplicates. To filter out the recording errors in Datastream, we
assign missing values to and 1 if (1 + )(1 + 1 ) 05 and at least one of them is
greater than or equal to 300%. is the stock return in month . For weekly and daily data, we
use 200% as the cut-o instead. In addition, in view of Datastreams practice to set the return
index to a constant once a stock ceases trading, we treat those constant values as missing values
in the inactive file.
In the first step, we determine domestic factors for each country. The domestic market factor
is given by the excess return in U.S. dollars of the countrys equity index return over the U.S.
Treasury Bill rate. We calculate country index returns using the MSCI country market indices
obtained from Datastream. For the size and book-to-market factors, we follow a methodology
similar to Fama and French (1993). All stocks reporting a market capitalization at the end of
June and a positive book-to-market ratio are double-sorted into two size groups and three bookto-market classifications. Half of the stocks are classified as large-cap () and the other half
as small-cap (). For the book-to-market classification, the bottom 30% of firms are classified
as , the middle 40% as , and the highest 30% as . The intersection of the rankings allows
for six value-weighted portfolios: and . Formally, we define
1
1
= ( + + ) ( + + )
3
3
1
1
= ( + ) ( + )
2
2
The monthly returns for and are then calculated from July in one year to June in
the next. The momentum factor () is constructed on a monthly basis. We rank stocks
at the end of month 1 based on their cumulative returns from 13 to 2 (i.e., prior 212
month returns by skipping month 1) and market value at the end of 1. Stock inclusion
in the portfolio construction requires nonmissing values for the cumulative return and market
value. For the market-cap classification, half of the stocks are again classified as large-cap
28

() and the other half as small-cap (). For the past returns classification, the bottom 30%
are classified as (low return), the middle 40% as , and the highest 30% as . The
momentum factor is defined as
1
1
= ( + ) ( + )
2
2
For the U.S. factors, we use the data posted on Kenneth R. Frenchs website. If a country has
fewer than 50 stocks qualifying for the portfolio construction, we set , , and
factors as missing for the respective year.
In the second step, we calculate a countrys international factors as the weighted average
domestic factors of all other countries. The weights are given by the relative stock market
capitalization of each country at the beginning of the year. The stock market capitalization
data is obtained from the World Development Indicator.
To obtain estimates of expected returns during the crisis period, we first estimate the loadings of each stock on the domestic and international risk factors ( = ) using a
regression over 60 months, from July 2002 to June 2007, as follows:
= +

1 + 2 + 3 + 4 +

where denotes a stocks monthly (cum dividend) return in U.S. dollars net of the one-month
Treasury Bill rate. For the pre-crisis period, July 2002 to June 2007, the average factor loadings
on the market, size, and value factors are positive. A negative average loading is found only
for the momentum factor. Untabulated results show that all eight factors have nonnegligible
explanatory power for the cross-section of returns. This is consistent with the recent evidence
by Eun et al. (2010), Hou, Karolyi, and Kho (2011), and Karolyi and Wu (2012) on the
importance of both local and global factors in stock returns.
b , the expected return for stock in month during
With the estimated factor loadings

the crisis period, July 2007December 2009, can be calculated as

=

b2 +
b3 +
b4
b1 +

29

References
[1] Amihud, Y., 2002, Illiquidity and stock returns: Cross-section and time-series eects,
Journal of Financial Markets 5, 3156.
[2] Aragon, G., and P. Strahan, 2012, Hedge funds as liquidity providers: Evidence from the
Lehman bankruptcy, Journal of Financial Economics 103, 570587.
[3] Badertscher, B. A., J. J. Burks, and P. D. Easton, 2012, A convenient scapegoat: Fair
value accounting by commercial banks during the financial crisis, Accounting Review 87,
5990.
[4] Bailey, W., A. Kumar, and D. Ng, 2011, Behavioral biases of mutual fund investors,
Journal of Financial Economics 102, 127.
[5] Bartram, S., J. Grin, and D. Ng, 2012, How important are foreign ownership linkages for international stock returns? Working Paper. Available at SSRN:
http://ssrn.com/abstract=1787425.
[6] Bekaert, G., M. Ehrmann, M. Fratzscher, and A. Mehl, 2012, Global crises and
equity market contagion, NBER Working paper no. 17121. Available at SSRN:
http://ssrn.com/abstract=2023372.
[7] Broner, F. A., R. G. Gelos, and C. Reinhart, 2006, When in peril, retrench: Testing the
portfolio channel of contagion, Journal of International Economics 69, 203230.
[8] Caballero, R., and A. Simsek, 2011, Fire sales in a model of complexity,
MIT Department of Economics Working Paper No. 09-28. Available at SSRN:
http://ssrn.com/abstract=1496592.
[9] Carhart, M., 1997, On persistence in mutual fund performance, Journal of Finance 52,
5782.
[10] Calomiris, C., I. Love, and M. S. Martinez Peria, 2012, Stock returns sensitivities to crisis
shocks: Evidence from developed and emerging markets, Journal of International Money
and Finance 31, 743765.
[11] Chan, W. S., R. Frankel, and S. P. Kothari, 2004, Testing behavioral finance theories using
trends and consistency in financial performance, Journal of Accounting and Economics 38,
350.
[12] Chevalier, J., and G. Ellison, 1997, Risk taking by mutual funds as a response to incentives,
Journal of Political Economy 105, 11671200.
[13] Coval, J., and E. Staord, 2007, Asset fire sales (and purchases) in equity markets, Journal
of Financial Economics 86, 479512.
[14] Covrig, V. M., M. L. Defond, and M. Hung, 2007, Home bias, foreign mutual fund holdings,
and the voluntary adoption of international accounting standards, Journal of Accounting
Research 45, 4170.
30

[15] Daniel, K., 2004, Discussion of: Testing behavioral finance theories using trends and
sequences in financial performance, (by Wesley Chan, Richard Frankel, and S.P. Kothari),
Journal of Accounting and Economics 38, 5164.
[16] Dornbusch, R., Y. Park, and S. Claessens, 2000, Contagion: Understanding how it spreads,
The World Bank Research Observer 15, 177197.
[17] Eun, C. , S. Lai, F. de Roon, and Z. Zhang, 2010, International diversification with factor
funds, Management Science 56, 15001518.
[18] Fama, E., and K. French, 1993, Common risk factors in the returns on stocks and bonds,
Journal of Financial Economics 33, 356.
[19] Fama, E., and K. French, 2010, Luck versus skill in the cross-section of mutual fund
returns, Journal of Finance 65, 19151947.
[20] French, K., 2008, Presidential address: The cost of active investing, Journal of Finance
63, 15371574.
[21] Forbes, K., and R. Rigobon, 2002, No contagion, only interdependence: Measuring stock
market comovements, Journal of Finance 57, 22232261.
[22] Frazzini, A., 2006, The disposition eect and underreaction to news, Journal of Finance
61, 20172046.
[23] Gorton, G., 2008, The panic of 2007, Yale ICF Working Paper No. 08-24. Available at
SSRN: http://ssrn.com/abstract=1255362.
[24] Gromb, D., and D. Vayanos, 2010, Limits of arbitrage: The state of the theory, Annual
Review of Financial Economics 2, 251-275.
[25] Hau, H., and H. Rey, 2008, Home bias at the fund level, American Economic Review:
Papers & Proceedings 98(2), 333338.
[26] Hau, H., and S. Lai, 2013, Real eects of stock underpricing, Journal of Financial Economics 108, 392408.
[27] Hou, K., G. A. Karolyi., and B. C. Kho, 2011, What factors drive global stock returns?,
Review of Financial Studies 24, 25272574.
[28] Kaminsky, G., C. Reinhart, and C. Vegh, 2003, The unholy trinity of financial contagion,
Journal of Economic Perspectives 17, 5174.
[29] Karolyi, G. A., and Y. Wu, 2012, The role of investability restrictions on size, value,
and momentum in international stock returns, Johnson School Research Paper Series No.
12-2012. Available at SSRN: http://ssrn.com/abstract=2043156.
[30] Kashyap, A., 2014, Open Questions for Monetary and Financial Stability Policy, Remarks
at the 2014 American Economic Association meeting.

31

[31] Kindleberger, C., 1978, Manias, Panics, and Crashes: A History of Financial Crises, Basic
Books, New York, NY.
[32] Kothari, S.P., 2001, Capital market research in accounting, Journal of Accounting and
Economics 31, 105231.
[33] Kumar, A., and C. Lee, 2006, Retail investor sentiment and return comovements, Journal
of Finance 61(5), 24512486.
[34] Lee, C. M. C., 2001, Capital market research in accounting: A discussion of capital market
research in accounting by S.P. Kothari, Journal of Accounting and Economics 31, 233253.
[35] Longsta, F., 2010, The subprime credit crisis and contagion in financial markets, Journal
of Financial Economics 97, 436450.
[36] Manconi, A., M. Massa, and A. Yasuda, 2012, The role of institutional investors in propagating the crisis of 20072008, Journal of Financial Economics 104, 491-518.
[37] Meier, I., and E. Schaumburg, 2006, Do funds window dress? Evidence for U.S. equity
mutual funds, Working Paper. Northwestern University.
[38] ONeal, E., 2001, Window dressing and equity mutual funds, Working Paper. Available at
SSRN: http://ssrn.com/abstract=275031.
[39] Rinne, K., and M. Suominen, 2010, Short term reversals, returns to liquidity provision and the costs of immediacy, working paper. Available at SSRN:
http://ssrn.com/abstract=1537923.
[40] Pulvino, T., 1998, Do asset fire sales exist? An empirical investigation of commercial
aircraft transactions, Journal of Finance 53, 939978.
[41] Seida, J. A., and W. F. Wempe, 2000, Do capital gain tax rate increases aect individual
investors trading decisions? Journal of Accounting and Economics 30, 3357.
[42] Shin, H. S., 2013, The Second Phase of Global Liquidity and Its Impact on Emerging
Economies, Keynote address at the Federal Reserve Bank of San Francisco Asia Economic
Policy Conference, November 3-5, 2013.
[43] Sias, R., 2006, Window dressing, tax-loss selling, and momentum profit seasonality, Financial Analysts Journal 62, 48-54.
[44] Sirri, E., and P. Tufano, 1998, Costly search and mutual fund flows, Journal of Finance
53, 15891622.

32

Table 1: Summary Statistics on Regression Variables

Reported are the summary statistics for all U.S. nonfinancial stocks. Fund exposure, is measured by the return
loss of a fund due to investment in financial stocks over the one-year period from July 1, 2007 to June 30, 2008. Stock
exposure, measures the ownership-weighted average exposure of all funds owning the stock. The dummy variable
marks with 1 the 30% of U.S. stocks with the highest stock exposure. We also define a separate measure of
stock exposure (20072) and the corresponding dummy variable (20072), which take into account fund
losses in financial stocks for just the second semester of 2007. Fund share, , denotes the aggregate ownership of
all funds in stock relative to the stocks market capitalization in June 2007. The dummy variable marks with
1 the 30% of U.S. stocks with the largest fund share. Cumulative risk-adjusted returns, () denote the return from
July 1, 2007, to the stated month or semesters later. The risk adjustment is based on an eight-factor international
asset pricing model with factor loadings estimated for the five-year pre-crisis period, July 2002June 2007. Percentage
e () denotes the aggregate change (over semesters) of all fund positions in
change in aggregate fund holdings
stock , relative to the aggregate positions in June 2007, multiplied by 100.
Variable

Obs.

Mean

Median

STD

Min

Max

Fund Exposure Measure

22 621

0026

0022

0027

0455

0000

Stock Exposure Measures

(20072) 100
100
(20072)

4 663
4 663
4 663
4 663

0116
0268
0300
0300

0076
0183
0000
0000

0128
0292
0458
0458

0975
2485
0000
0000

0000
0000
1000
1000

Fund Share Measures

4 663
4 663

0171
0300

0140
0000

0156
0458

0000
0000

0882
1000

Cumulative Risk-Adjusted
Stock Returns
(1) (Dec. 2007)
(2) (June 2008)
(3) (Dec. 2008)
(4) (June 2009)
(5) (Dec. 2009)

3 691
3 599
3 497
3 389
3 179

0080
0085
0107
0017
0070

0029
0087
0265
0296
0327

0607
0839
1483
1112
1065

0822
0934
0993
0992
0996

3246
5100
14211
8349
9329

(Nov. 7, 2008)
(Feb. 27, 2009)

3 449
3 411

0302
0378

0141
0226

1801
2285

0987
0994

16381
22411

Percentage Change in
Aggregate Fund Holdings
e (1) (Dec. 2007)

e (2) (June 2008)

e (3) (Dec. 2008)

e (4) (June 2009)

e (5) (Dec. 2009)

4 203
4 221
4 177
4 170
4 119

2563
4786
6400
7406
8845

1042
2528
3801
4745
6738

4212
6401
7715
8500
8903

24003
33187
36681
38425
39999

5229
5681
5258
5621
3879

33

Table 2: Dierences between Exposed and Nonexposed Stocks

Exposed stocks are the 30% of U.S. nonfinancial stocks with fund owners experiencing the largest return losses due
to investments in financial stocks over the one-year period from July 2007 to June 2008, and nonexposed stocks are
the remaining 70% of stocks. Panel A reports the mean and median of exposed and nonexposed stocks for stock
capitalization in the natural logarithm of U.S. dollars (LnSize), Amihud illiquidity measure (Illiquidity), the price-tobook ratio, the receivables-to-sales ratio, leverage (or the total debt-to-asset ratio), the short-term (ST) debt-to-asset
ratio, and dividend yield. The market capitalization and the price-to-book ratio are based on the data in June 2007
from Datastream. The receivables-to-sales ratio, leverage, the short-term (ST) debt-to-asset ratio, and dividend yield
are based on the latest fiscal year-end data prior to July 2007 obtained from the Compustat database. Panel B reports
the distribution of the number of funds holding an exposed stock (Columns 13) and a nonexposed stock (Columns
46) in June 2007. We distinguish between exposed and nonexposed fund owners. Fund exposure is measured by the
return loss of a fund due to ownership in financial stocks over the one-year period from July 2007 to June 2008. The
30% of funds with the largest exposure to the financial sector are marked as exposed funds, and the remaining 70% as
nonexposed funds.

Panel A: Firm Characteristics

Exposed Stocks
Mean
Median

Variable

Obs.

LnSize
Illiquidity
Receivables-to-Sales
Price-to-Book
Leverage
ST Debt-to-Assets
Dividend Yield

1 399
1 350
1 340
1 345
1 350
1 365
1 367

21238
0037
0159
3063
0212
0027
0010

21120
0001
0149
2326
0196
0005
0000

Nonexposed Stocks
Obs.
Mean
Median
3 264
2 392
2 143
2 539
2 183
2 225
2 259

18553
0890
0203
3890
0168
0031
0009

18709
0034
0154
2659
0082
0003
0000

Panel B: Fund Ownership Distribution

Exposed Stocks
All
Exp.
Non-Exp.
Fund
Fund
Fund
Owners Owners
Owners
Percentile
p1
p5
p10
p25
p50
p75
p90
p95
p99
N
Mean
SD

Non-Exposed Stocks
All
Exp.
Non-Exp.
Fund
Fund
Fund
Owners Owners
Owners

11
41
71
126
189
297
476
616
996

5
15
22
42
64
118
272
348
561

6
22
44
77
116
163
230
285
475

1
1
1
5
21
68
138
199
397

0
0
0
1
7
18
37
52
181

0
1
1
3
13
49
100
150
259

1 394
241
1911

1 394
109
1153

1 394
131
907

2 796
52
802

2 796
15
286

2 796
37
559

34

Table 3: OLS Regressions for Cumulative Stock Returns

The cumulative risk-adjusted stock returns (starting from July 1, 2007) over one to five consecutive semesters are
regressed on two dummy variables. The dummy variable marks with 1 the 30% of U.S. stocks with fund owners
most exposed to financial stocks. A fund owners exposure is measured by its return loss in financial stocks from July 1,
2007 to June 30, 2008. In the first regression (Column 1), the contemporaneous stock exposure dummy (20072)
takes into account fund owners return loss in financial stocks for only the second semester of 2007. The dummy variable
marks with 1 the 30% of U.S. stocks with the largest overall fund ownership in June 2007. The last two columns
regress the cumulative weekly stock returns from June 29, 2007 to the twin peaks of the crisis (November 7, 2008 and
February 27, 2009) on the two dummy variables. All regressions include industry fixed eects. The t-values based on
robust standard errors are reported in brackets.

Dec. 2007
(20072)

Cumulative Risk-Adjusted Returns (by)

End of Semester
Peak of Crisis
Nov. 7, Feb. 27,
June 2008 Dec. 2008 June 2009 Dec. 2009
2008
2009

0116
[502]

0100
[413]

0153
[425]
0209
[559]

3 691
0023

3 599
0053

0206
[348]
0205
[329]

0096
[196]
0227
[456]

0033
[073]
0230
[478]

0315
[444]
0120
[161]

0409
[419]
0269
[273]

3 497
0013

3 389
0032

3 179
0035

3 449
0011

3 411
0023

35

Table 4: Quantile Cumulative Stock Return Regressions with Liquidity Controls

Reported are quantile regressions for the cumulative (weekly) U.S. stock returns starting from June 29, 2007 to November
7, 2008 and February 27, 2009. The dummy variable (marking the 30% of U.S. stocks with the highest exposure
to distressed funds) and the dummy variable (marking the 30% of U.S. stocks with the highest overall fund
ownership) are the same as those defined in Table 1. The dummy variable marks the 30% most liquid U.S.
stocks, based on the Amihud illiquidity measure. is the natural logarithm of stock capitalization value. The
explanatory power of the regression is reported for the 25%, 50%, 75%, and 90% quantiles of the cumulative stock
returns. All regressions include industry fixed eects. The t-values based on bootstrapped standard errors are reported
in brackets.

Liquidity Control
Quantile 25%

Cumulative Risk-Adjusted Returns (by)

Nov.7 Feb. 27
Nov.7 Feb. 27
Nov.7
Feb. 27
2008
2009
2008
2009
2008
2009

and
0041
[189]
0096
[248]
0193
[687]

0001
[002]
0077
[222]
0235
[787]

Quantile 50%

0038
[112]
0145
[364]
0151
[331]

0029
[048]
0143
[227]
0253
[625]

Quantile 75%

0232
[512]
0092
[133]
0048
[096]

0313
[376]
0237
[370]
0235
[304]

Quantile 90%

0694
[487]
0211
[140]
0274
[263]

0921
[437]
0219
[125]
0071
[041]

Q25%
Q50%
Q75%
Q90%

2
2
2
2

3 449
0058
0042
0036
0160

3 411
0050
0044
0042
0190

36

0005
[017]
0062
[222]

0030
[069]
0067
[180]

0010
[034]
0051
[197]
0047
[153]
0056
[707]

0038
[137]
0052
[225]
0104
[221]
0044
[762]

0062
[1119]

0060
[1237]

0110
[247]
0143
[401]

0039
[080]
0131
[348]

0107
[245]
0131
[340]
0042
[098]
0046
[584]

0077
[157]
0130
[293]
0129
[233]
0042
[459]

0051
[898]

0059
[621]

0222
[258]
0107
[224]

0301
[417]
0231
[285]

0219
[362]
0097
[140]
0070
[111]
0008
[033]

0325
[391]
0225
[319]
0185
[170]
0021
[102]

0002
[016]

0047
[230]

0667
[317]
0171
[119]

0700
[445]
0212
[112]
0061
[144]

0649
[386]
0201
[160]
0134
[078]
0069
[132]

0808
[504]
0196
[109]
0235
[130]
0082
[160]

0085
[268]
3 449
0064
0048
0038
0157

3 411
0056
0050
0044
0190

3 449
0059
0043
0037
0167

3 411
0050
0044
0042
0191

Table 5: Cumulative Stock Returns and Firm Characteristics

This table repeats the baseline results in Table 3, with additional controls for various firm characteristics, including
the Amihud illiquidity measure (Illiquidity), the receivables-to-sales ratio, the price-to-book ratio, leverage (the total
debt-to-asset ratio), the short-term (ST) debt-to-asset ratio, and dividend yield, measured based on the latest fiscal
year-end data prior to July 2007.

Cumulative Risk-Adjusted Returns (by)

End of Semester
Dec. 2007
(20072)

June 2008

Dec. 2008

June 2009

Dec. 2009

0087
[344]

0145
[381]
0179
[463]

0183
[290]
0201
[308]

0073
[144]
0233
[449]

0010
[020]
0235
[466]

0263
[358]
0154
[197]

0359
[341]
0332
[327]

0002
[030]
0000
[027]
0000
[144]
0171
[218]
0314
[146]
0225
[087]

0002
[044]
0001
[046]
0000
[027]
0278
[239]
0534
[198]
0390
[130]

0004
[050]
0002
[077]
0001
[165]
0305
[183]
0884
[214]
0503
[207]

0010
[288]
0001
[022]
0003
[338]
0288
[205]
0359
[091]
0081
[036]

0006
[135]
0005
[235]
0000
[103]
0211
[165]
0769
[265]
0024
[012]

0004
[037]
0002
[064]
0002
[196]
0285
[152]
0910
[202]
0679
[182]

0014
[113]
0002
[034]
0003
[198]
0311
[127]
0550
[075]
0523
[134]

2 783
0041

2 711
0091

2 643
0035

2 591
0063

2 495
0083

2 639
0041

2 616
0041

0091
[366]

Illiquidity
Receivables-to-Sales
Price-to-Book
Leverage
ST Debt-to-Asset
Dividend Yield

Peak of Crisis
Nov. 7, Feb. 27,
2008
2009

37

Table 6: OLS Regressions for Aggregate Fund Holding Changes

For each stock, the percentage change in the aggregate fund holdings relative to positions in June 2007 over one to four
consecutive semesters is regressed on dummy variables. The dummy variable marks with 1 the 30% of U.S.
stocks with fund owners most exposed to financial stocks. A fund owners exposure is measured by its return loss in
financial stocks from July 1, 2007 to June 30, 2008. In the first set of regressions (Columns 12), the contemporaneous
stock exposure dummy (20072) takes into account fund owners return loss in financial stocks for only the
second semester of 2007. The dummy variable marks the 30% of U.S. stocks with the largest fund ownership
share in June 2007. The dummy marks the 30% of U.S. stocks with the highest cumulative return over the
semester(s) under consideration. The dummy represents the interaction of the stock exposure
dummy (or (20072)) and the high crisis-return dummy All regressions include industry fixed
eects. The t-values based on robust standard errors are reported in brackets.

Dec. 2007
(20072)

0848
[408]

2531
[1191]

June 2009

0843
[397]

Percentage Change in Aggregate Fund Holdings

June 2008
Dec. 2008

2508
[1179]

1612
[500]

1488
[448]

2158
[597]

1772
[470]

2697
[707]

2159
[516]

5415
[1646]

5364
[1646]

7491
[2045]

7515
[2089]

8608
[2233]

8639
[2269]

0507
[359]

1080
[573]

1246
[589]

1161
[521]

0341
[099]

0917
[199]

1235
[240]

1330
[247]

4 203
0150

4 203
0154

4 221
0259

4 221
0269

38

4 177
0326

4 177
0338

4 170
0352

4 170
0361

Table 7: Structural VAR Estimates

The daily MSCI U.S. market index return is combined in Panel A with the daily return of a long-short DMF
(direct minus fund) portfolio and in Panel B with the daily return of a long-short RMI (retail minus institutional)
portfolio, in order to identify the panic eect of index return shocks on stocks with a high share of direct retail
investors. The structural VAR is just identified under the restriction that there are no contemporaneous intra-day
eects from the index return to the long-short portfolio return (i.e., 21 = 0), which is based on the assumption that
retail investors react with at least one day of delay to index shocks. In the equation of in Panel A, the regression

coecients (1) and (1) capture the eect of 1

and 1
on . Other regression coecients
are defined analogously. The pre-crisis period, crisis period I, and crisis period II refer to 01/07/200630/06/2007,
01/07/200730/06/2008, and 01/07/200830/06/2009, respectively. The detailed VAR specification is described in
Section 4.5.
Panel A: Market Index and DMF Portfolio
Pre-Crisis Period
Coecient
z -value

Crisis Period I
Coecient
z -value

Crisis Period II
Coecient
z -value

Equation for
0
(1)
(2)
(1)
(2)
Joint significance

0001
0067
0054
0057
0018
2 (5) = 162

[200]
[074]
[060]
[040]
[013]
= 0805

0001
0230
0003
0219
0132
2 (5) = 989

[096]
[294]
[004]
[144]
[094]
= 0042

0001
0148
0217
0176
0047
2 (5) = 1578

[067]
[218]
[297]
[094]
[027]
= 0003

Equation for
0
(1)
(2)
(1)
(2)
Joint significance

0000
0104
0033
0078
0112
2 (5) = 562

[173]
[184]
[058]
[087]
[126]
= 0230

0000
0294
0120
0138
0183
2 (5) = 7251

[032]
[743]
[275]
[179]
[258]
= 0000

0001
0158
0117
0023
0139
2 (5) = 6659

[153]
[659]
[452]
[035]
[223]
= 0000

1123
0004
0004

[1588]
[2227]
[2227]

1182
0010
0006

[1179]
[2236]
[2236]

1061
0026
0010

[641]
[2236]
[2236]

12
11
22

248

250

250

Panel B: Market Index and RMI Portfolio

Pre-Crisis Period
Coecient
z -value

Crisis Period I
Coecient
z -value

Crisis Period II
Coecient
z -value

Equation for
0
(1)
(2)
(1)
(2)
Joint significance

0001
0001
0019
0223
0120
2 (5) = 695

[216]
[002]
[029]
[203]
[109]
= 0139

0000
0103
0060
0500
0005
2 (5) = 1689

[062]
[158]
[092]
[321]
[003]
= 0002

0001
0239
0260
0228
0154
2 (5) = 1800

[066]
[313]
[338]
[144]
[102]
= 0001

Equation for
0
(1)
(2)
(1)
(2)
Joint significance

0000
0067
0036
0006
0100
2 (5) = 874

[107]
[162]
[087]
[009]
[148]
= 0068

0000
0119
0090
0096
0032
2 (5) = 2692

[047]
[440]
[335]
[149]
[051]
= 0000

0000
0113
0023
0055
0037
2 (5) = 1138

[014]
[303]
[061]
[071]
[050]
= 0023

0585
0006
0004

[607]
[2227]
[2227]

0611
0012
0005

[412]
[2236]
[2236]

1252
0022
0014

[1230]
[2236]
[2236]

12
11
22

248

250

39

250

Table 8: International Evidence

The return regressions of Table 3 are repeated for nonfinancial stocks in the international markets. Panels A, B, and
C report the results for all non-U.S. stocks, developed market stocks excluding the U.S. stocks, and emerging market
stocks, respectively. The regressions use equal country weights, and all stocks are given equal weight within a country.

Panel A: All non-U.S. Stocks

Dec. 2007
(20072)

Cumulative Risk-Adjusted Returns (by)

End of Semester
Peak of Crisis
Nov. 7, Feb. 27,
June 2008 Dec. 2008 June 2009 Dec. 2009
2008
2009

0031
[253]

0024
[192]

0076
[363]
0007
[033]

14 691
0206

14 666
0269

0134
[284]
0054
[112]

0025
[079]
0002
[006]

0019
[050]
0022
[058]

0168
[310]
0065
[115]

0235
[316]
0133
[174]

14 651
0253

14 631
0232

14 608
0212

14 622
0228

14 618
0235

Panel B: Developed Market Stocks ex U.S.

Dec. 2007
(20072)

Cumulative Risk-Adjusted Returns (by)

End of Semester
Peak of Crisis
Nov. 7, Feb. 29,
June 2008 Dec. 2008 June 2009 Dec. 2009
2008
2009

0047
[318]

0004
[029]

0085
[321]
0017
[064]

9 969
0165

9 938
0170

0134
[212]
0115
[171]

0029
[067]
0037
[080]

0026
[048]
0046
[089]

0168
[228]
0135
[172]

0261
[258]
0275
[258]

9 921
0200

9 904
0199

9 882
0162

9 900
0173

9 897
0191

Panel C: Emerging Market Stocks

Dec. 2007
(20072)

Cumulative Risk-Adjusted Returns (by)

End of Semester
Peak of Crisis
Nov. 7, Feb. 29,
June 2008 Dec. 2008 June 2009 Dec. 2009
2008
2009

0014
[068]

0045
[217]

0062
[182]
0009
[028]

4 722
0238

4 728
0367

0124
[184]
0037
[057]

0014
[030]
0051
[107]

0006
[012]
0015
[027]

0157
[207]
0041
[056]

0175
[170]
0083
[082]

4 730
0313

4 727
0283

4 726
0314

4 722
0287

4 721
0283

40

Table 9: Abnormal Pre-Crisis Return Dierence: Exposed vs. Nonexposed Stocks and Funds
Reported are the risk-adjusted return dierences (i) between exposed and nonexposed stocks and (ii) between exposed
and nonexposed funds, prior to the crisis. Exposed stocks are the 30% of U.S. nonfinancial stocks with fund owners
experiencing the largest return losses due to investments in financial stocks over the one-year period from July 2007 to
June 2008, and nonexposed stocks are the remaining 70%. Similarly, the 30% of funds with the largest exposure to the
financial sector are marked as exposed funds and the remaining 70% as nonexposed funds. Following Fama and French
(2010), we form an equal-weighted portfolio and a value-weighted portfolio, separately, for the exposed and nonexposed
stocks each month from January 2002 to December 2006. We then test for their dierences in risk-adjusted returns,
controlling for the standard risk factors in the literature (i.e., the market, size, book-to-market, and momentum factors)
and allowing the risk factor loadings to dier across the two types of stocks. Using a similar approach, we also form
value- and equal-weighted portfolios of the U.S. nonfinancial stock holdings of, separately, exposed and nonexposed
funds. The dierence in regression intercept () and the associated standard deviation and T -value, as well as the
adjusted 2 of the regression, are reported.

Exposed vs. Nonexposed Stocks

Equal Weighted Portfolio
Value Weighted Portfolio
Exposed vs. Nonexposed Funds
Equal Weighted Portfolio
Value Weighted Portfolio

Di. in

Std. Dev.

T -value

00044
00028

00047
00028

093
099

0926
0954

00005
00016

00019
00019

026
085

0977
0976

41

Fire Sale Effects for Exposed Stocks

-.1
-.2
-.3

01
n2
ja

ju

01

01

01

9
l2

00
n2
ja

ju
01

00

8
l2

00
n2

-.1
-.2
-.3
-.4

Cumulative Returns

01
n2
ja

01

ju
01

n2
ja

l2

00

00

8
00
01

01

ju

l2

00
n2
ja

01

01

01

ja

ju

n2

l2

01

00

-.5
0

9
ju

01

n2
ja
01

l2

00

00

8
00
l2
ju

01

n2
ja
01

ju

l2

00

00

-.5

-.4

-.3

-.2

-.1

Cumulative Returns

D. Emerging Markets

C. Developed Markets ex U.S.

01

00

7
01

01

ja

ju

n2

l2

01

00

9
ja
01

01

01

ja

ju

n2

l2

00

00

8
00
l2
ju

01

01

01

ja

ju

n2

l2

00

00

-.5

-.4

Cumulative Returns

-.1
-.2
-.3
-.4
-.5

Cumulative Returns

B. Non-U.S. Stocks

A. U.S. Stocks

Figure 1: The graph shows the underperformance of exposed, nonfinancial stocks relative to other nonfinancial stocks
in the same industry after accounting for risk premia from a model with four local and four international risk factors.
Exposed stocks are the 30% of nonfinancial stocks with their fund owners most exposed to the financial sector during
the one-year period from July 2007 to June 2008. Panels A, B, C, and D plot the graph for, respectively, U.S. stocks,
all non-U.S. stocks, developed market stocks excluding the U.S., and emerging market stocks. The vertical bars provide
robust standard errors (1 standard deviation) around the point estimate of the average cumulative risk-adjusted
returns.

42

Nonexposed Stocks

.05

.1

.15

.2

Density

.05

.1

.15

.2

Exposed Stocks

-40

-20

20

Percentage Aggregate Fund Holding Change

Graphs by stock_type

Figure 2: Plotted is the distribution of the percentage change in the aggregate fund holdings over four consecutive
e (4). Exposed stocks are the 30% of U.S. nonfinancial stocks with
semesters relative to positions in June 2007, i.e.,
fund owners experiencing the largest return losses due to investments in financial stocks over the one-year period from
July 2007 to June 2008, and nonexposed stocks are the remaining 70%.

43

.6
.4
.2
0
-.2
0

Crisis Period I
1/7/2007 - 30/6/2008
.6
.4
.2
0
-.2
0

Trading Days

Cumulative IFR of DMF Portfolio

Pre-Crisis Period
1/7/2006 - 30/6/2007

Cumulative IFR of DMF Portfolio

Cumulative IRF of DMF Portfolio

Panel A: Impulse Response of DMF Portfolio to Market Return Shock

Crisis Period II
1/7/2008 - 30/6/2009
.6
.4
.2
0
-.2
0

Trading Days

5
Trading Days

.3
.2
.1
0
-.1
0

5
Trading Days

Crisis Period I
1/7/2007 - 30/6/2008

Cumulative IFR of RMI Portfolio

Pre-Crisis Period
1/7/2006 - 30/6/2007

Cumulative IFR of RMI Portfolio

Cumulative IRF of RMI Portfolio

Panel B: Impulse Response of RMI Portfolio to Market Return Shock

.3
.2
.1
0
-.1
0

5
Trading Days

Crisis Period II
1/7/2008 - 30/6/2009
.3
.2
.1
0
-.1
0

5
Trading Days

Figure 3: In Panel A, we estimated a VAR consisting of the daily MSCI U.S. market return index and a value-weighted
long-short DMF (direct minus fund) portfolio that longs the 30% of U.S. stocks with the lowest fund ownership and
shorts the 30% with the highest fund ownership in June 2007. In Panel B, we replace the DMF portfolio with the
RMI (retail minus institutional) portfolio that longs the 30% of NYSE stocks with the highest share of retail trading
volume and shorts the 30% with the lowest share. Plotted are the cumulative impulse response functions (IRFs) for
the DMF and RMI portfolio returns after a unit innovation to the U.S. index return for three separate time periods:
The pre-crisis period, crisis period I, and crisis period II. The upper and lower lines provide a 95% confidence interval
for the point estimates.

44

Stock Exposure Distribution by Markets

B. Developed Markets ex U.S.

C. Emerging Markets

20

Percent

40

60

A. U.S.

-.01

-.005

-.01

-.005

-.01

-.005

Stock Exposure
Graphs by country_type

Figure 4: Plotted are the distributions of stock exposure for nonfinancial stocks in the U.S. (Panel A), developed
markets excluding the U.S. (Panel B), and emerging markets (Panel C). Stocks with less than 001 of stock exposure
account for only a small proportion of the population and are therefore not plotted.

45

Distribution of Pairwise Fund Portfolio Overlap

Portfolio Overlap
.2 .4 .6 .8

Pairs of Funds in Group A

20

40

60

80

100

80

100

Percentile

Portfolio Overlap
.2 .4 .6 .8

Pairs of Funds in Group B

20

40

60
Percentile

Portfolio Overlap
0 .2 .4 .6 .8

Pairs of One Fund from Group A and one Fund from Group B

20

40

60

80

100

Percentile

Figure 5: Group A consists of the 10% most distressed funds (i.e., funds with the greatest investment loss in financial
stocks over the one-year period from July 2007 to June 2008). We then match this group of funds with the same number
of other funds, based on their country codes and the size of their total asset holdings in nonfinancial stocks. The group
of matched funds is labeled as Group B. The portfolio overlap statistic is then calculated for (i) pairs of funds in Group
A (upper panel), (ii) pairs of funds in Group B (middle panel), and (ii) pairs of one fund from Group A and one fund
from Group B (bottom panel), based on fund holdings in December 2006.

46