Preliminary and incomplete. Comments welcome. Please do not cite or circulate without permission.

Do Fire Sales Create Externalities?*

Sergey Chernenko sergey.chernenko@fisher.osu.edu

The Ohio State University

Adi Sunderam asunderam@hbs.edu

Harvard University and NBER

May 1, 2017

Abstract

We construct three novel measures of much of the price impact of their trading different mutual

funds are likely to internalize. We show that high internalization funds hold larger cash buffers

and use these buffers more aggressively to accommodate fund inflows and outflows. As a result,

stocks held by high internalization funds experience lower volatility, and flows into high

internalization funds have smaller spillover effects on other funds holding the same securities.

Our results suggest that there are significant fire sale externalities in the mutual fund industry,

and that a planner coordinating among funds would choose different liquidity management

policies.

* We are grateful to Carter Anthony, Jules van Binsbergen, Charles Calomiris, Jaewon Choi, Lauren Cohen, Andrew Ellul, Robin Greenwood, Johan Hombert, Zoran Ivkovic, Marcin Kacperczyk, Xuewen Liu, Stefano Rossi, Alexi Savov, Jeremy Stein, René Stulz, Robert Turley, Jeff Wang, Zhi Wang, Michael Weisbach, and Yao Zeng for helpful comments and suggestions. Some of the results in this paper were previously included in a working paper by the same authors titled “Liquidity Transformation in Asset Management: Evidence from the Cash Holdings of Mutual Funds.”

1

Fire sales are at the center of our understanding of financial stability problems in

securities markets. The key dynamic in models of fire sales is that sales by one fund tighten

constraints on other funds leading to further sales (e.g., Geanakoplos, 2009; Stein, 2012). For

instance, redemptions from an open-end mutual fund can force sales of the securities the fund

holds. This depresses the prices of those securities, thereby stimulating redemptions at other

funds holding the same securities due the performance-flow relationship. These redemptions lead

to further fire sales, thus amplifying the initial shock. The key feature of most models of fire

sales is that they typically generate externalities: a planner coordinating among funds would

chose different actions, dampening the feedback effects across funds.

A large empirical literature on fire sales documents the key input ingredients of the

theory. Forced sales do result in depressed security prices (e.g., Coval and Stafford, 2007, Ellul,

Jotikasthira and Lundblad (2011), Greenwood and Thesmar (2011), and Merrill et al (2012)).

Furthermore, these depressed prices do appear to have adverse effects on other funds holding the

same securities (Lou (2012), Falato et al (2016)). However, the empirical literature has not

documented the main result of the theoretical models on fire sales: that there are indeed

externalities so that a planner coordinating among funds would chose different actions. This

conclusion is also not obvious given the existing literature. It could well be the case that the

probability of fire sales or the costs of spillovers across funds are low enough that a planner

would not choose a different outcome from the private market equilibrium.

In this paper, we aim to fill the gap in the literature. Motivated by theory, we construct

three novel measures of how much of the price impact of their trading different mutual funds are

likely to internalize. We then relate these measures to the flow management behaviors of funds,

the securities they hold, and spillovers to other funds. We find that funds that internalize more of

the price impact of their trading do behave substantially differently from those that do not,

suggesting that there are indeed significant fire sale externalities in the mutual fund industry.

Our analysis uses a novel data set on the cash holdings of equity mutual funds collected

from the SEC form N-SAR filings. Importantly, our data set covers holdings of both cash and

cash substitutes such as money market mutual fund shares. In recent years, cash substitutes have

become an increasingly important source of liquidity for asset managers. For instance, in 2014,

2

mutual funds held $600 billion of cash and cash substitutes, with nearly 50% taking the form of

cash substitutes.

We begin by proposing three new measures of internalization for mutual funds. The first

one is based upon the intuition that a monopolist internalizes its price impact. Indeed, in most

theories of fire sales, the planner’s problem is essentially the problem a monopolist would face.

This suggests that funds that are more like monopolists in the securities they hold should

internalize more of their price impact. Our second measure leverages the fact that many

managers manage multiple funds and thus presumably care about exerting adverse price pressure

across funds. Our third measure exploits the idea that a fund may be cautious about exerting

price impact when it may adversely affect other funds in the same fund family, even if those

funds are not managed by the same portfolio manager.

Using these measures, we present four main results. Throughout our analysis, we use all

three of our internalization measures. Even though these measures have low correlations with

each other, all three deliver similar results. Our first result concerns how funds’ management of

the inflows and outflows they receive varies with our internalization measures. We find that

funds that internalize their price impact more use their cash buffers more aggressively to

accommodate flows. The economic magnitudes are significant. A fund with a one standard

deviation higher internalization measure accommodates one dollar of flows using 11-16% more

cash, and a corresponding amount less trading in its underlying portfolio securities, than the

average fund. This is similar to the impact of the fund having assets that are one standard

deviation more illiquid. This suggests that there is an externality: a planner would have funds

follow a different strategy for accommodating inflows and outflows, just as the funds that are

higher on our internalization measures do.

These results are robust to a variety of controls for alternative explanations. Specifically,

we provide evidence that our results cannot be explained by market timing on the part of

individual funds, asset liquidity, fund strategy, variation in investor clienteles across funds, and

manager characteristics. In addition, we get similar results when we split the sample into small

and large funds or split the sample before and after the financial crisis.

Our second main result is at the security level. We show that when an individual security

is held by funds that internalize more of their price impact, its realized volatility is lower over the

3

following quarter. Magnitudes here are modest in absolute terms, but significant relative to the

total amount of security-level volatility induced by fund flows. A one-standard deviation

increase in our internalization measures increases security-level volatility as much as a one-

standard deviation increase the total volatility induced by fund flows, as measured by the

Greenwood and Thesmar (2011) measure of fragility. This suggests that the price impact

externality funds impose on one another has meaningful consequences for the behavior of asset

prices.

Our third main result concerns the degree to which flows into one fund impact the

performance of other funds holding the same securities. We show that funds that internalize

more of the price pressure from their trading do indeed exert a smaller effect on the returns of

other funds. Specifically, we show that when the securities one fund holds are held by other

funds that internalize more of their price pressure, the relationship between the first fund’s

returns and flows into the other funds is diminished. The economic magnitudes are again

meaningful: a one-standard deviation increase in our internalization measures reduces the

relationship between returns and flows into funds with overlapping holdings by 8-11%.

Finally, we examine the relation between our internalization measures and the amount of

cash funds hold. We find that funds with higher internalization measures hold larger cash

buffers. The magnitudes are economically significant, again comparable to the variation in the

size of cash buffers induced by variation in asset liquidity. Our other results are ex post: they

concern how funds deal with inflow and outflows once they receive them. This result is

essentially ex ante: it shows that funds that internalize more of their price impact in fact behave

differently before any flows have been realized. In some sense, this is closest to the problem a

planner would solve in a theoretical model.

Overall, our results provide strong evidence that fire sale externalities operate across

equity mutual funds. A planner coordinating between funds would have them act in a

substantially different way than they do in the private market equilibrium. It is important to note

that our results do not imply a welfare statement. For there to be a social loss from low cash

4

holdings in general equilibrium, the liquidation costs to the funds must not simply be a transfer

to an outside liquidity provider.1

Our paper is related to several strands of the literature. First, there is a large theoretical

and empirical literature studying fire sales in debt and equity markets, including Shleifer and

Vishny (1992), Shleifer and Vishny (1997), Coval and Stafford (2007), Ellul, Jotikasthira and

Lundblad (2011), Greenwood and Thesmar (2011), and Merrill et al (2012).2 Our results show

how mutual funds use cash holdings to manage the risk of fire sales created by their liquidity

transformation activities and suggest that they may not hold enough cash to fully mitigate fire

sale externalities.

In addition, we contribute to a small but growing literature on the determinants and

effects of mutual fund cash holdings, including Yan (2006), Simutin (2014), and Hanouna,

Novak, Riley, and Stahel (2015). While this literature focuses primarily on funds’ market timing

ability and the impact cash holdings have on returns, we use cash holdings along with our

internalization measures to empirically measure the extent to which funds internalize the price

impact they exert on security prices.

The remainder of the paper is organized as follows. Section I outlines our empirical tests.

Section II describes the data. Section III presents our main results, and Section IV concludes.

I. Framework

In this section, we outline the economic logic behind our measures of internalization and

our main empirical tests.

A. Measures of Internalization

1 For example, fire sales that result in banks suffering losses may impair the functioning of the bank lending channel (see, for example, Bernanke and Blinder (1988), Kashyap and Stein (2000), Stein (2010). Similarly, losses and redemptions by high yield mutual funds can negatively affect the investment of speculative-grade firms (Chernenko and Sunderam (2012)). Even highly rated firms that borrow in money markets may find it difficult to immediately substitute to other sources of financing when money market mutual funds experience large redemptions (Chernenko and Sunderam (2014)). 2 In addition, there is a broader literature on debt and equity market liquidity, including Roll (1984), Amihud and Mendelsohn (1986), Chordia, Roll, and Subrahmanyam (2001), Amihud (2002), Longstaff (2004), Acharya and Pedersen (2005), Bao, Pan, and Wang (2011), Dick-Nielsen, Feldhütter, and Lando (2012), Feldhütter (2012), and many others.

5

Throughout the paper, we use three different measures of internalization. We describe the

construction of these variables in more detail in Section II below. Here we discuss the economic

motivation for examining each variable. The first one is Share outstanding - the weighted-

average of a fund’s holdings of each portfolio security relative to the security’s outstanding

amount. The idea here is that fire sale externalities stem from the fact that multiple funds hold

the same security. When any individual fund sells the security, it does not internalize the effect

that has on the other funds. This logic implies that a monopolist in a particular security

internalizes its price impact. When a monopolist creates price impact through trading, it is the

only fund that suffers because it is the only one that holds the security. Our Share outstanding

variable generalizes this intuition: the higher it is, the closer to a monopolist the fund is in the

securities it holds.

Our second measure of internalization leverages the fact that many mutual fund managers

manage more than one fund. Our Manager overlap variable measures the overlap in portfolio

holdings across multiple funds managed by a given portfolio manager. The idea here is that

managers care about the joint performance of all the funds they manage. Thus, they will be

reluctant to trade if the price impact generated by the trading of one of their funds adversely

affects the performance of another one of their funds. The scope for these adverse impacts is

greater when portfolio holdings overlap a lot across the funds the manager manages.

Our third measure of internalization is based on the idea that fund families may at least

partially internalize price impact across different funds in the family. This kind of internalization

is sometimes incentivized by the compensation contracts of fund managers.3 For instance,

according to the Prospectus of Metropolitan West Funds, “Many portfolio managers participate

in equity incentives based on overall firm performance of the TCW Group and its affiliates,

through ownership or participation in restricted unit plans that vest over time or unit appreciation

plans of the Adviser's parent company.” Similarly, the Prospectus of Oppenheimer Rising

Dividends Fund states, “the long-term award component consists of grants in the form of

appreciation rights in regard to the common stock of the Sub-Advisers holding company parent,

restricted shares of such common stock, as well as deferred cash investments in the fund(s)

managed by a portfolio manager.” Thus, if a fund holds assets that are also held by other funds in

3 Ibert et al (2017) show the importance of firm-level revenues and profit for the compensation of mutual fund managers in Sweden.

6

the same family, then the fund may be more likely to internalize the price impact of its trading on

those funds than on funds outside the fund family. Our Family overlap measure captures the

overlap in portfolio holdings across multiple funds in the same fund family.

B. Empirical Predictions

We next outline the main empirical predictions we test in the body of the paper. The first

prediction concerns how mutual funds respond to inflows and outflows from their investors.

Prediction 1. Funds that internalize more of their price impact will use their cash holdings

more aggressively to accommodate inflows and outflows.

The idea here is that funds that internalize more of their price impact will be more reluctant to

immediately trade in their portfolio securities because doing so would exert significant price

pressure. Instead, they will use cash buffers to manage inflows and outflows, trading more

slowly in their portfolio securities to minimize price impact over time.

Our next two predictions explore the implications of these differences in flow

management practices across funds. The second prediction concerns the effects of this flow

management behavior on security prices.

Prediction 2. Securities held by funds that internalize more of their price impact will have

lower realized volatility.

The idea here is that if funds that internalize their price impact more are successful at using their

cash buffers to manage inflows and outflows, the volatility of the securities they hold should be

lower.

The third prediction concerns the effect of flow management on other funds holding the

same securities.

Prediction 3. The relationship between a fund’s returns and the flows of other funds holding

the same securities will be weaker when the other funds internalize more of their price impact.

The idea here is that price pressure from flow management varies with how much funds

internalize their price impact. In general, if funds holding the same securities as fund f suffer

7

outflows, their trading will tend to push down fund f’s returns. However, if those funds

internalize more of their price impact, the relationship between their flows and f’s returns will be

weaker.

Our final prediction is ex ante. Prior to receiving inflows or outflows, do funds that

internalize their price impact more behave differently?

Prediction 4. Funds that internalize more of their price impact will hold more cash.

The idea here is that to avoid exerting too much price impact through trading, funds that

internalize their price impact will hold large cash buffers ex ante.

II. Data

A. Cash holdings

We combine novel data on the cash holdings of open-end mutual funds with several other

data sets. Our primary data comes from the SEC form N-SAR filings. These forms are filed

semi-annually by all mutual funds and, among other things, provide data on asset composition,

including holdings of cash and cash substitutes. Specifically, we measure holdings of cash and

cash substitutes as the sum of cash (item 74A), repurchase agreements (74B), short-term debt

securities other than repurchase agreements (74C), and other investments (74I). Short-term debt

securities have remaining maturities of less than a year and consist mostly of US Treasury Bills

and commercial paper.

The other investments category (74I) consists mostly of investments in money market

mutual funds (MMMFs), other mutual funds, loan participations, and physical commodities.

Using hand-collected data, we have examined the composition of the other investments category

for a random sample of 320 funds for which other investments accounted for at least 10% of total

net assets. The mean and median fractions of MMMFs in other investments were 75% and

100%. Holdings of other mutual funds accounted for most of the remaining value of other

investments. We use our security-level holdings data, described below, to subtract holdings of

long-term mutual funds from other investments. Otherwise, we treat the other investments

8

category as consisting entirely of MMMFs. This should only introduce measurement error into

our dependent variable and potentially inflate our standard errors.4

Cash holdings reported on a fund’s balance include cash collateral received when lending

out portfolio securities. This cash collateral is usually legally segregated and is not available to

meet redemption requests. We use a Python script to extract the amount of securities lending

collateral from N-CSR filings, and subtract securities lending collateral from the gross value of

cash holdings reported in N-SAR filings.

Our dependent variable is thus the sum of cash and cash equivalents, net of securities

lending collateral, scaled by TNA (item 74T). We winsorize this cash-to-assets ratio at the 1st

and 99th percentiles.

In addition to data on asset composition, form N-SAR contains data on fund flows and

investment practices. Gross and net fund flows for each month since the last semi-annual filing

are reported in item 28. Item 70 reports indicators for whether the fund uses various types of

derivatives, borrows, lends out it securities, or engages in short sales.5

B. Link to CRSP mutual fund database

For additional fund characteristics such as investment objective, fraction of institutional

share classes, and holdings liquidity, we link our N-SAR data to the CRSP Mutual Fund

Database. Using a name-matching algorithm, we can match the majority of funds in N-SAR to

CRSP.6 We match more than 70% of all fund-time observations in N-SAR to CRSP. In dollar

terms, we match more than 80% of all assets. 4 The CRSP Mutual Fund Database includes a variable called per_cash that is supposed to report the fraction of the fund’s portfolio invested in cash and equivalents. This variable appears to be a rather noisy proxy for the cash-to-assets ratio. Aggregate cash holdings of all long-term mutual funds in CRSP track aggregate holdings of liquid assets of long-term mutual funds as reported by the Investment Company Institute (ICI) until 2007, but the relationship breaks down after that. By 2014, there is a gap of more than $400 billion, or more than 50% of the aggregate cash holdings reported by ICI. At a more granular level, we calculated cash holdings from the bottom up using security-level data from the SEC form N-CSR for a random sample of 100 funds. The correlation between the true value of the cash-to-assets ratio computed using N-CSR data and our N-SAR based proxy is 0.75. The correlation between the true value and CRSP is only 0.40. 5 Almazan et al (2004) also use form N-SAR’s investment practices data. 6 Our procedure takes advantage of the structure of fund names in CRSP. The full fund name in CRSP is generally of the form “trust name: fund name; share class.” For example, “Vanguard Index Funds: Vanguard 500 Index Fund; Admiral Shares.” The first piece, “Vanguard Index Funds,” is the name of the legal trust that offers Vanguard 500 Index Fund as well as a number of other funds. Vanguard Index Funds is the legal entity that files on behalf of Vanguard 500 Index Fund with the SEC. The second piece, “Vanguard 500 Index Fund,” is the name of the fund itself. The final piece, “Admiral Shares,” indicates different share classes that are claims on the same portfolio but that offer different bundles of fees, minimum investment requirements, sales loads, and other restrictions.

9

After linking our data to CRSP, we apply the following screens to our sample of funds.

We focus on open-end funds and exclude exchange-traded funds (ETFs),7 small business

investment companies (SBIC), unit investment trusts (UIT), variable annuities, funds of funds,8

and money market mutual funds. In addition, we exclude observations with zero assets according

to N-SAR and those for which the financial statements do not cover a regular 6- or 12-month

reporting period. As we discuss below, we are able to measure asset liquidity for domestic equity

funds, identified using CRSP objective codes starting with ED.9 To further make sure that we can

accurately measure fund flow volatility and asset liquidity, we focus on funds with at least $100

million in assets. Finally, we exclude index funds for two reasons. First, index funds are likely to

have higher carrying costs (i.e., costs of tracking error) than other funds. Thus, for index funds,

cash holdings are likely to be lower and less sensitive to asset liquidity and fund flow volatility.

Second, index funds largely track the most liquid securities, so there is little variation in asset

liquidity for us to analyze among them.

C. Asset liquidity

We use holdings data from the CRSP Mutual Fund Database to measure the liquidity of

equity mutual fund holdings.10 These data start in 2003. Following Chen, Goldstein, and Jiang

(2010), we construct the square root version of the Amihud (2002) liquidity measure for each

stock. We then aggregate up to the fund level, taking the value-weighted average of individual

stock liquidity.

7 ETFs operate a very different model of liquidity transformation. They rely on investors to provide liquidity in the secondary market for the fund’s share and on authorized participants (APs) to maintain parity between the market price of the fund’s shares and their NAV. In untabulated results, we find that ETFs hold significantly less cash and that to the extent that they do hold more than a token amount of cash, it is almost entirely due to securities lending and derivatives trading. 8 SBICs, UITs, and open-end funds are identified based on N-SAR items 5, 6, and 27. ETFs are identified based on the ETF dummy in CRSP or fund name including the words ETF, exchange-traded, iShares, or PowerShares. Variable annuities are identified based on N-SAR item 58. We use security-level data from CRSP and Morningstar to calculate the share of the portfolio invested in other mutual funds. Funds that, on average, invest more than 80% of their portfolio in other funds are considered to be funds of funds. 9 Corporate bond funds are defined as funds that have Lipper objective codes A, BBB, HY, IID, MSI, and MSI and that invest more than 50% of their portfolio in intermediate and long-term corporate bonds (NSAR item 62P). 10 In unreported analyses, we obtain very similar results when we use Thomson Reuters Mutual Funds Holdings data.

10

D. Internalization

Using CRSP Mutual Fund holding data, we construct three measures of internalization.

The first one is Share outstanding - the weighted-average of a fund’s holdings of each portfolio

security relative to the security’s outstanding amount:

𝑆ℎ𝑎𝑟𝑒 𝑜𝑢𝑡𝑠𝑡𝑎𝑛𝑑𝑖𝑛𝑔! =𝑉!,!𝑉!,!!

×𝑉!,!

𝑂𝑢𝑡𝑠𝑡𝑎𝑛𝑑𝑖𝑛𝑔!!

where f indexes funds and s indexes securities. Because data on the amount outstanding comes

from CRSP, calculation of Share outstanding is limited to portfolio securities with valid

PERMNOs and ignores holdings of non-equity or private securities.11 Because our sample

consists of domestic equity funds, omission of securities not in CRSP constitutes a small fraction

of fund holdings.

Our second measure of internalization is Manager overlap, which measures the overlap in portfolio holdings across multiple funds managed by a given portfolio manager. For each security s held by fund f, we calculate aggregate holdings of the security by all other funds managed by the same portfolio manager and divide this by the aggregate TNA of all funds managed by the portfolio manager. We then calculate the value-weighted average across all securities held by fund f. Specifically,

𝑀𝑎𝑛𝑎𝑔𝑒𝑟 𝑜𝑣𝑒𝑟𝑙𝑎𝑝! =𝑉!,!𝑉!,!!

×𝑉!,!!"# ! !!"# ! ,!!!

𝑇𝑁𝐴!!"# ! !!"# !!

Portfolio manager identity comes from Morningstar. For funds managed by multiple portfolio managers we split total holdings of each security equally among the fund’s managers.

The third measure, Family overlap, is meant to capture the overlap in portfolio holdings

across multiple funds in the same fund family. Even if such funds are not managed by the same

individual portfolio managers, the fund family may provide incentives for portfolio managers to

at least partially internalize the impact of their trading decisions on the performance of other

funds within the family. Formally,

11 Chernenko et al (2017) show for example that some US equity mutual funds have been increasing their investment in unicorns: private firms with valuations of at least $1 billion. For the average fund however holdings of unicorns still account for only 2 basis points of TNA.

11

𝐹𝑎𝑚𝑖𝑙𝑦 𝑜𝑣𝑒𝑟𝑙𝑎𝑝! =𝑉!,!𝑉!,!!

×𝑉!,!!"#$%& ! !!"#$%& ! ,!!!

𝑇𝑁𝐴!!"#!"# ! !!"#$%& !!

Note that since Manager overlap and Family overlap do not use data on securities’ outstanding amounts, we can construct these measures using all portfolio securities (with valid CUSIPs) and not restrict the calculation to stocks in CRSP.

E. Summary statistics

Our final data set is a semi-annual fund-level panel that combines the N-SAR data with

additional fund information from CRSP and data on holdings liquidity and internalization from

CRSP and Morningstar. We conduct most of our analyses at the fund-half year level. The sample

period, determined by the availability of holdings data in CRSP, is January 2003 – June 2016.

Table 1 reports basic summary statistics for funds in our data. Our sample of equity funds

consists of about 22,500 observations.12 Median fund has TNA of about $600 million of which

about 2.73% is held in cash and equivalents. There is significant variation across funds in their

cash-to-assets ratio: the interquartile range is 1.15%-5.15%. Consistent with an aggregate shift

towards index funds, the median fund in our data experiences quarterly fund flows of -0.5%.

Turning to the internalization measures, the typical fund owns 0.15% of its securities and

has Manager and Family overlap of 0.20% and 0.12%. with significant cross-sectional variation

across funds.

Appendix Table A1 gives formal definitions for the construction of all variables used in

the analysis.

III. Results

We now present our tests of the key empirical predictions outlined above. It is worth

noting that for much of the analysis, we are documenting endogenous relationships. Fund

characteristics, investor behavior, and cash holdings are all jointly determined, and our results

trace out the endogenous relationships between them.13

12 The number of bond funds in our sample is significantly smaller than the number of equity funds because we focus on bond funds that invest at least 50% of their portfolio in corporate bonds. 13 In most cases, endogeneity should lead to coefficients that are smaller in magnitude. For instance, Chen, Goldstein, and Jiang (2010) argue that higher cash holdings should endogenously lower the volatility of fund flows

12

A. Liquidity management through cash holdings

We begin by understanding the baseline flow management behavior of mutual funds,

showing that cash holdings play an important role in the way mutual funds manage inflows and

outflows. In Table 2, we estimate regressions of the change in a fund’s cash holdings over the

last six months on the net flows it received during each of those six months:

ΔCashf ,m−6:m =αobj ( f ),m +β0Flowsf ,m + ...+β5Flowsf ,m−5 +ε f ,m. (6)

Fund flows are winsorized at the 1st and 99th percentiles; results are similar when winsorizing at

the 5th and 9th percentiles.

In the first three columns, the dependent variable is the change in cash holdings over the

last six months as a fraction of net assets six months ago: ΔCashf ,m−6:m /TNAf ,m−6 . The coefficient

β0 = 0.17 is large and highly statistically significant. Since flows are scaled by the same

denominator – assets six months ago – as the dependent variable, the coefficients can be

interpreted as dollars. Thus, β0 = 0.17 indicates that a dollar of outflows during month m

decreases cash holdings by 17 cents. Similarly, a dollar of inflows increases cash holdings by 17

cents. The other 83 cents are met by transacting in the fund’s holdings of equities.14 The

specification includes Lipper objective code cross time (half-year) fixed effects, indicating that

the results are not driven by relationships between flows and cash holdings in particular fund

objectives.

The coefficient β0 shows that an economically significant portion of flows is

accommodated through cash holdings. Even though equities are quite liquid, and a month is a

relatively long period, 17% of flows at a monthly horizon are accommodated through changes in

cash holdings. Presumably, at higher frequencies (e.g., daily or weekly), cash plays an even more

important role. The remaining coefficients show that the effect of fund flows on cash holdings

declines over time. By month m-3 the coefficient is 0.029. Since the median cash-to-assets ratio

is 2.73%, this coefficient is consistent with funds scaling their portfolio up and down.

because investors are less worried about fire sales. This should weaken the relationship between cash and fund flow volatility relative to the case where fund flow volatility is exogenous. 14 These results are broadly consistent with Edelen (1999), who finds that a dollar of fund flows is associated with about 70 cents in trading activity.

13

The second column of Table 2 shows that the results are similar if we include fund fixed

effects, showing that the results are not driven by a correlation between the average change in

cash and the average flows an individual fund faces over the sample period. Finally, the third

column shows that the results remain essentially unchanged if we include both fund fixed effects

and Lipper objective code cross time fixed effects.

In the last three columns of Table 2, the dependent variable is the change in the fund’s

cash-to-assets ratio:

ΔCashTNA

⎛

⎝⎜

⎞

⎠⎟f ,m

=CashTNA

⎛

⎝⎜

⎞

⎠⎟f ,m

−CashTNA

⎛

⎝⎜

⎞

⎠⎟f ,t−6

.

These regressions show that funds are not simply responding to flows by scaling their portfolios

up and down. The overall composition of the portfolio is changing, becoming more cash-heavy

when a fund receives inflows and less cash-heavy when it suffers outflows.

The coefficient β0 = 0.068 is statistically and economically significant. Flows equal to

100% of assets increase the fund’s cash-to-assets ratio by 6.8% (percentage points). For

reference, the standard deviation of fund flows is 9%. Note that the coefficients here are likely to

be biased down because of the performance-flow relationship. If a fund has strong returns

between month m-6 and month m, it is likely to receive inflows, but its cash-to-assets ratio at

time m will be depressed because high returns inflate assets at time m.

The remaining columns in Table 2 show that the results are robust to the inclusion of

fund and Lipper objective code cross time fixed effects.

Appendix Table A2 estimates equation (6) splitting fund flows into subscriptions versus

redemptions. We find that funds respond relatively symmetrically to inflows and outflows. This

is consistent with the idea that funds care about the price pressure they exert on the underlying

assets when both buying and selling.

B. Internalization and Flow Management

We next examine how the propensity to use cash for flow management varies with our

internalization measures. Table 3 estimates specifications that allow flow management practices

to differ across the cross section of funds based on our internalization measures:

14

ΔCashf ,t−2:t =αobj ( f ),t +β1Flowsf ,t +β2Flowsf ,t × Internalize f ,t−2 +β3Flowsf ,t × Illiq f ,t−2+β4Flowsf ,t−1 +β5Flowsf ,t−1 × Internalize f ,t−2 +β6Flowsf ,t−1 × Illiq f ,t−2+β7Internalize f ,t−2 +β8Illiq f ,t−2 +ε f ,t .

(7)

For compactness, we aggregate flows into quarters, and refer to time in quarters.15 We interact

quarterly flows with the lagged values of our internalization measures. Thus, the specification

asks: does the way the fund responds to fund flows vary with how much we expect it to

internalize its price impact?

Equation (7) controls for quarterly flows interacted with the illiquidity of the fund’s

holdings, as measured by the value-weighted average of the square root version of the Amihud

(2002) measure for each of fund’s holdings. This helps us separate liquidity from internalization.

Liquidity is a security-level concept: a fund with less liquid securities should use cash more

aggressively because the fund itself incurs transaction costs when it trades those securities.

Internalization is a holdings-level concept: it is determined by which funds hold the security and

how much they care about exerting price impact on one another.

In the first three columns, the dependent variable is the change in cash holdings over the

last two quarters as a fraction of assets two quarters ago: ΔCashf ,t−2:t /TNAf ,t−2 . We standardize

the internalization and illiquidity variables so that their coefficients can be interpreted as the

effect of a one-standard deviation change in each variable. Again, all specifications include

Lipper objective cross time fixed effects.

The first column of Table 3 shows that for the average equity fund, one dollar of flows

over the most recent quarter t changes cash holdings by β1 = 15 cents. For a fund that internalizes

its price impact one standard deviation more than the average fund, as measured by

ShareOutstanding , one dollar of flows changes cash holdings by β1 + β2 = 17.4 cents, a 16%

larger effect. In comparison, for a fund with assets one standard deviation less liquid than the

average fund, the same dollar of flows changes cash holdings by 17.1 cents. Thus, the impact of

our internalization measures is economically sizeable. The reduction in trading we would

observe if funds internalized more of their price impact from trading is similar to the reduction

we would observe if their assets were substantially less liquid.

15 Interacting monthly flows with asset illiquidity generates somewhat stronger results for more recent fund flows.

15

The second and third columns of Table 3 show that similar results obtain using our other

measures of internalization. A fund that internalizes its price impact one standard deviation more

than the average fund, as measured byManagerOverlap , is 12% more aggressive in using cash

to manage flows than the average fund. A fund that internalizes its price impact one standard

deviation more than the average fund, as measured by FamilyOverlap , is 11% more aggressive

in using cash to manage flows than the average fund. Despite the fact that our measures are not

highly correlated with each other, we obtain similar results with all of them. Funds that

internalize more of their price impact are more aggressive in using cash to manage inflows and

outflows.

The last three columns of Table 3 use the change in the fund’s cash-to-assets ratio as the

dependent variable. Here we generally find slightly larger effects. In the fourth column, a 100%

quarterly flow increases the cash-assets ratio of the average fund by 5.5 percentage points. For a

fund that internalizes its price impact one standard deviation more than the average fund, as

measured by ShareOutstanding , the cash-assets ratio increases by 6.9 percentage points, 25%

more. In comparison, a one-standard deviation increase in the illiquidity of the fund’s holdings

increases the sensitivity of cash to flows by 15%. Similar results hold in the fifth and sixth

columns, where we use ManagerOverlap and FamilyOverlap , respectively, as our

internalization measures.

C. Robustness of Flows Results

The results in Table 3 show that our measures of internalization strongly affect funds’

flow management strategies. Funds that score highly on our internalization measures use cash

more aggressively to meet inflows and outflows. This is consistent with the idea that they care

more about minimizing the price impact of their trades. However, given that our internalization

measures are indirect proxies, there could be other explanations for these results.

Table 4 reports a battery of robustness tests that examine these alternative explanations.

Each row of the table shows the results a different robustness test. Across the columns, we report

the results using both ΔCashf ,t−2:t /TNAf ,t−2 and the change in the cash-assets ratio as the

dependent variable and using each of our internalization measures. All specifications include

fund objective cross time fixed effects. Row (1) replicates our baseline results from of Table 3.

16

In Row (2) of Table 4, we control for the fund’s monthly returns between each of our

observations of cash holdings. Because of the performance-flow relationship, returns and flows

are correlated. When a fund has high returns, its asset base increases, depressing its cash-assets

ratio at the same time it is likely to receive inflows and thus biasing down the coefficient on

flows. If funds with high internalization measures have weaker performance-flow relationships,

this could explain the results in Table 3. Row (2) of Table 4 shows that controlling for past

returns has no impact on our results.

Whether funds accommodate fund flows through changes in cash or trading in the

underlying securities could depend on fund managers’ expectations of future returns. Row (3) of

Table 4 uses future returns realized over the following 1, 3, 6, and 12 months as proxies for

expected returns. Controlling for these does not affect the coefficient of interest.

We next turn to the idea that our internalization measures are just additional proxies for

the illiquidity of fund holdings. Though we control for the interaction of flows and illiquidity in

Table 3, the effect of illiquidity may be non-linear. Rows (4), (5), and (6) of Table 4 explore this

alternative. Row (4) shows that controlling for five powers of illiquidity and their interactions

with flows has no effect on our results. Row (5) shows that the same conclusion holds when we

control for deciles of illiquidity and their interactions with flows.16

In Row (6) we examine the possibility that the weighted average liquidity of a portfolio

may not fully capture the effect of holdings illiquidity on flow management. Consider two funds

– A and B - whose holdings have the same average liquidity. Suppose all securities held by fund

A have the same liquidity, while half of securities held by fund B are more liquid than average

while the other half is less liquid. Even though the two funds have the same average liquidity,

fund B may trade its more liquid securities and may avoid adjusting its cash holdings in response

to fund flows. A negative correlation between our internalization measures and dispersion in

liquidity across securities held by a given fund could then explain the positive coefficient on the

interaction of fund flows and internalization. To rule out this possibility, Row (6) of Table 4

separately controls for the liquidity of each 10% slice of portfolio from most to least liquid (and

the interaction of each of these with fund flows). Overall, Rows (4), (5), and (6) of Table 4 are

16 Deciles of illiquidity are defined relative to the full sample of fund-half year observations.

17

strong evidence that our internalization measures are not simply proxies for the illiquidity of

fund holdings.

We next consider the idea that our internalization measures could be correlated with fund

strategies. For instance, having a high value of ShareOutstanding could indicate that the fund

manager tends to make big bets. We explore this alternative in Rows (7), (8), and (9) of the

Table 4. In Row (7) we control for the concentration of the fund’s holdings, as measured by the

Herfindahl-Hirschmann Index (HHI) of portfolio weights, and its interaction with fund flows.

This does not affect our results. The same is true in Row (8), where we control for the share of

the single largest position in the fund’s portfolio and its interaction with flows. In Row (9), we

control for the fund’s active share as defined by Cremers and Petajisto (2009) and its interaction

with flows. The point estimates on the interaction between our internalization measures and

flows here are bigger than in the baseline sample. However, the sample is much smaller, only

40% the size of the baseline sample, and thus the coefficients are not statistically significant in

two of six cases. Overall, Rows (7), (8), and (9) of the Table 4 provide strong evidence that our

results are not driven by differences in strategies across funds.

In Row (10), we consider the possibility that our internalization measures are correlated

with the investor clienteles that different funds serve. For instance, having more performance-

sensitive clients could change a fund’s flow management strategy. We control for the number of

share classes, the concentration of fund assets across share classes, and whether the fund charges

a front load and their interaction with flows. These controls have no effect on our results.

In Row (11), we consider the possibility that our internalization measures reflect

differences in fund manager characteristics. For example, more experienced portfolio managers

tend to manage more funds and to have greater overlap in their holdings. It could be that instead

of capturing the effect of internalization our results are capturing differences in the behavior of

more versus less experienced managers. To rule out this possibility, we control for whether the

fund is team managed, the number of managers, the years of experience of the manager, whether

the manager has a CFA, and the interactions of these variables with fund flows. Again this has

no effect on our main results.

18

Finally, in Rows (12)-(15) we examine two sample splits. First, Rows (12) and (13) show

that we get similar results for small (below median TNA) and large (above median) funds. In

Rows (14) and (15) we examine the impact of our internalization measures during the first and

second half of our sample period. Given the smaller sample size, the standard errors here are

larger and statistical significance is weaker. Overall, however, we find broadly similar results

during the two subperiods.

Overall, Table 4 provides strong evidence that we are picking up the impact of

internalization on flow management. Our examination of alternative explanations for our results

does not seem consistent with those explanations driving our results.17

D. Internalization and Fragility

We next explore the implications of differences in flow management practices across

funds for the price behavior of the assets that funds hold. Specifically, we ask whether the stocks

held by funds that internalize more of their price impact experience lower realized volatility. One

might expect this to be the case because such funds are less likely to trade in their holdings right

away in response to flows. To test this prediction, we estimate the regression

Vols,t+1 =αs +αt+1 +β ⋅ Internalizes,t + ʹγ X s,t +εs,t+1

where Vols,t+1 is the realized volatility of stock s in quarter t+1 and Internalizes,t is the average of

one of our internalization measures across the holders of s at the end of quarter t, weighted by the

size of their holdings. The regression includes stock and time fixed effects. In addition, we

include in the set of controls X s,t the size (log market capitalization) of the stock, the fraction of

the total market capitalization held by mutual funds, and the fragility measure of Greenwood and

Thesmar (2011). This measure is meant to capture volatility induced by mutual fund flows across

all the holders of the stock.18

17 It is worth noting that many of the alternative explanations we examine do affect how funds manage their flows. However, controlling for these alternative explanations does not affect our key result that internalization is associated with greater propensity to accommodate fund flows through cash holdings. 18 We use the “diagonal” version of Greenwood and Thesmar (2011) fragility measure that ignores the correlation in fund flows across funds. Column (4) of Table 3 in Greenwood and Thesmar (2011) shows that the diagonal version of fragility generates similar results to the full version that accounts for cross-correlation.

19

Table 5 reports the results, with the dependent variable expressed in percentage form. In

the first column, we use ShareOutstanding as our internalization measure. A one-standard

deviation increase in the average level of internalization across holders of a given stock reduces

that stock’s volatility by 0.39 percentage points. Given that mean volatility for the stocks in our

sample is 38.7%, this is not a very large effect in absolute terms. However, the relevant

benchmark here is not overall volatility but excess volatility induced by mutual fund trading.

Internalization cannot affect the part of a stock’s volatility that is generated by news about

fundamentals, it can only affect volatility induced by trading. Thus, it is more natural to compare

the effect of internalization to the effect of the Greenwood and Thesmar (2011) fragility

measure. As the first column of Table 7 shows, a one-standard deviation increase in fragility is

associated with a 0.18 percentage point increase in volatility. Compared to this benchmark, the

effect of internalization is significant.

In the second and third columns of Table 5, we split stocks into liquid (below median

illiquidity) and illiquid (above median). As expected, the effects of internalization (and of

fragility) are stronger for less liquid stocks. In these stocks, the reluctance of funds that

internalize their price impact to trade has a larger impact on volatility.

The remaining columns of Table 5 show that we get similar results for our other

internalization measures. A one standard deviation increase in ManagerOverlap is associated

with a 0.23 percentage point decrease in volatility, while a one standard deviation increase in

FamilyOverlap is associated with a 0.18 percentage point decrease. These magnitudes are again

similar to the impact of flow-induced mutual fund trading overall, as measured by fragility. And

again, the effects are stronger for illiquid stocks.

In summary, Table 5 shows strong evidence that when funds that internalize their price

impact hold the same stock that stock’s volatility is lower.

E. Internalization and Spillovers on Fund Performance

In most theoretical models of fire sales, forced sales by some mutual funds exert negative

price pressure, thereby driving down the returns of other mutual funds, which because of the

performance-flow relation are in turn forced to liquidate some of their holdings. How do our

internalization measure affect the relation between fund returns and flows into other funds with

20

overlapping holdings? Specifically, consider a particular fund f. When other funds that hold the

stocks that f holds have outflows, their trading in those stocks may drive down prices and thus

fund f’s returns. However, to the extent that those other funds internalize more of the price

pressure from their trading, the effect on f’s returns will be smaller. To examine this prediction,

we estimate the regression

Rf ,m =αm +β1 ⋅Pressure f ,m +β2 ⋅Pressure f ,m × Internalize f ,m−1+β3 ⋅ Internalize f ,m−1 +β4 ⋅Flowsf ,m +ε f ,m

where Rf ,m is the return of fund f during month m. Pressure f ,m is the average over securities that

fund f holds of the flows into other funds that hold each security, weighted by share of each

security in f’s portfolio and the share of each other fund in total holdings of the security.

Internalize f ,m is the average over securities that fund f holds of the internalization of other funds

that hold each security with the same weighting scheme. Because index funds have no scope to

accommodate fund flows through changes in cash holdings, we set their value of internalization

to zero. In addition, we control directly for the flows into fund f and include time fixed effects in

all specifications.

Table 6 shows the results. In the first column, we measure internalization using

ShareOutstanding . The coefficient 1β on Pressure is positive: when funds that have holdings

overlapping with f have outflows, fund f has lower returns.19 However, the coefficient 2β on the

interaction Pressure× Internalize is negative. The effect is muted when those funds with

overlapping holdings internalize more of their price pressure. A one-standard deviation increase

in internalization decreases the effect of Pressure by 7%.

The second and third columns of Table 6 show that the results get a bit stronger when we

include fund fixed effects or objective cross time fixed effects. With fund fixed effects a one-

standard deviation increase in internalization decreases the effect of Pressure by 8%. With

objective cross time fixed effects a one-standard deviation increase in internalization decreases

the effect of Pressure by 10%.

19 Our results here are similar to Lou (2012). The main difference is that while he is interested in predicting mutual fund returns based on expected flows into other funds holding the same securities as a given fund, our focus here is on ex-post behavior: how does internalization moderate the effect of flow induced trading on other funds.

21

Columns (4)-(6) of Table 6 show that similar results obtain when we measure

internalization using FamilyOverlap . The economic magnitudes are also similar: a one-standard

deviation in internalization decreases the effect of Pressure by 9-11%.

F. Internalization and Cash Holdings

Our results so far are ex post: they show how funds manage fund flows once they receive

them and how flow management depends on fund’s propensity to internalize the price impact of

its trading. We now turn to the ex ante choice of cash buffers and show that funds that internalize

more of their price impact choose to hold higher cash buffers. In some sense, this is closest to the

problem a planner would solve in a theoretical model.

Table 7 estimates regressions linking the cash-assets ratio to our internalization measures:

Cashf ,tTNAf ,t

=α +β ⋅ Internalization f ,t + ʹγ X f ,t +ε f ,t .

Controls X f ,t include four sets of variables. The first set is variables that capture the mismatch

between liquidity of a fund’s assets and liquidity demanded by its investors: the liquidity of the

assets, the volatility of fund flows, and the interaction of the two. The second category consists

of regressors that capture economies of scale: the (log) size of the fund and the (log) size of the

fund family. In the third category is the fraction of the fund’s assets that are in institutional share

classes, a proxy for investor behavior. Finally, we control for a number of measures of trading

practices, including the fund’s asset turnover and indicators for whether the fund uses various

derivatives, borrows, lends out its securities, or engages in short sales.

The first column of Table 7 shows that our internalization measures have a substantial

impact on cash holdings. A fund that internalizes its price impact one standard deviation more

than the average fund, as measured by ShareOutstanding , has a 0.74 percentage point higher

cash-assets-ratio. In comparison, increasing the illiquidity of the fund’s assets by one standard

deviation raises the cash-assets ratio by 0.26 percentage points. Thus, as in our flow management

regressions, internalization has an economically significant effect. A fund internalizing more of

its price impact has as much cash as a fund with significantly less liquid assets.

22

The second column of Table 7 adds objective cross time fixed effects to the regression.

The results are unchanged indicating that they are not driven by common time variation in

internalization and cash holdings within fund objectives.

The remaining columns of Table 7 show that results are similar for our other

internalization measures, though the magnitudes are somewhat smaller. A fund that internalizes

its price impact one standard deviation more than the average fund, as measured by

ManagerOverlap , has a cash-assets ratio that is 0.2 percentage points higher than the average

fund. A fund that internalizes its price impact one standard deviation more than the average fund,

as measured by FamilyOverlap has a cash-assets ratio that is 0.25 percentage points higher than

the average fund. Though the magnitudes are smaller, they are still significant when compared to

the effect of increasing the illiquidity of the fund’s assets.

G. Robustness of Cash Holdings Results

In Table 7, we examine the robustness of our cash holdings results. We use the same

battery of tests we did in Table 4. The table shows that our results are robust to controls for

market timing (i.e., fund returns), asset liquidity, fund strategy, investor clienteles, and manager

characteristics. In addition, we get similar results when we split the sample into small and large

funds or split the sample period in two.

Overall, Table 7 provides strong evidence that we are picking up the impact of

internalization on ex ante cash holdings. Our examination of alternative explanations for our

results does not seem consistent with those explanations driving our results.

IV. Conclusion

Theoretical models of fire sales suggest that investors do not fully internalize the cost of

fire sales they create. They therefore may set their leverage too high or hold too little liquidity to

meet redemption requests. The existing empirical literature documents that forced sales do result

in depressed security prices, i.e., that fire sales do exist, and that these depressed prices do appear

to have adverse effects on other investors. However, there is no direct empirical evidence that the

costs of spillovers effects are significant enough that a planner coordinating among funds would

choose a different outcome from the private market equilibrium.

23

Motivated by theory, we construct three novel measures of how much of the price impact

of their trading different mutual funds are likely to internalize. While they are not strongly

correlated, all three measures deliver the same message—funds that internalize more of the price

impact of their trading behave differently from other funds. In particular funds high on our

internalization measures use cash more aggressively to accommodate fund flows. As a result,

stocks held by these funds experience less excess volatility, and flows into these funds have

smaller spillover effects on other funds with overlapping holdings. Funds high on our

internalization measures also choose to hold larger cash buffers ex ante. Overall, these results

suggest that there are indeed significant fire sale externalities in the mutual fund industry and

that, like funds with high values of internalization, a planner coordinating among funds would

choose different liquidity management policies.

Our results speak to the policy debate among academics, practitioners, and regulators

about whether liquidity transformation in asset management can cause financial stability

problems (e.g., Goldstein et al, 2015; International Monetary Fund, 2015; Financial Stability

Oversight Council, 2014; Feroli et al, 2014; Chen, Goldstein, and Jiang, 2010). Indeed,

concerned in part by the potential for fire sales, the Securities and Exchange Commission (SEC)

has recently proposed new rules to promote more effective liquidity risk management by mutual

funds (SEC (2015)). Our results suggest that because individual mutual funds do not fully

internalize the price impact of their trading, they are likely to hold less cash and to meet

redemption requests by trading more aggressively in the underlying securities than would a

planner coordinating among funds.

Finally, the results in this paper contribute to our understanding of the role of large

institutional investors in securities markets (Ben-David et al 2016). While larger investors have

the potential to generate greater price impact, they are also more likely to more fully internalize

the price impact of their trading. In our data, mutual funds that have greater overlap in holdings

with other funds in the same fund family hold larger cash buffers and use these buffers more

aggressively to accommodate fund flows. As a result stocks held by such funds experience lower

volatility, while other funds holding such stocks are more insulated from their fund flows.

24

References

Acharya, Viral, and Lasse Pedersen, 2005, Asset Pricing with Liquidity Risk, Journal of Financial Economics 77, 375-410.

Almazan, Andres, Keith C. Brown, Murray Carlson, and David A. Chapman, 2004, Why Constrain Your Mutual Fund Manager? Journal of Financial Economics 73, 289-321.

Amihud, Yakov, 2002, Illiquidity and Stock Returns: Cross-Section and Time Series Effects, Journal of Financial Markets 5, 31-56.

Amihud, Yakov, and Haim Mendelson, 1986, Liquidity and Stock Returns, Financial Analysts Journal 42, 43-48.

Bao, Jack, Jun Pan, and Jiang Wang, 2011, The Illiquidity of Corporate Bonds, Journal of Finance 66, 911–946.

Ben-David, Itzhak, Francesco Franzoni, Rabih Moussai, and John Sedunov, 2016, The Granular

Nature of Large Institutional Investors, NBER working paper 22247.

Chen, Qi, Itay Goldstein, and Wei Jiang, 2010, Payoff Complementarities and Financial Fragility: Evidence from Mutual Fund Outflows, Journal of Financial Economics 97, 239-62.

Chernenko, Sergey, Josh Lerner, and Yao Zeng, 2017, Mutual Funds as Venture Capitalists? Evidence from Unicorns, Unpublished working paper.

Chernenko, Sergey, and Adi Sunderam, 2014, Frictions in Shadow Banking: Evidence from the Lending Behavior of Money Market Mutual Funds, Review of Financial Studies 27(6), 1717-1750.

Chernenko, Sergey, and Adi Sunderam, 2012, The Real Consequences of Market Segmentation, Review of Financial Studies 25(7), 2041-2070.

Chevalier, Judith and Glenn Ellison, 1997, Risk Taking by Mutual Funds as a Response to Incentives, Journal of Political Economy 105, 1167-1200.

Chevalier, Judith and Glenn Ellison, 1999, Are some mutual fund managers better than others? Cross-sectional patterns in behavior and performance, Journal of Finance 54, 875-99.

Chordia, Tarun, Richard Roll, and Avanidhar Subrahmanyam, 2001, Market Liquidity and Trading Activity, Journal of Finance 56, 501-530.

Coval, Joshua, and Erik Stafford, 2007, Asset Fire Sales (and Purchases) in Equity Markets, Journal of Financial Economics 86, 479-512.

Dick-Nielsen, Jens, Peter Feldhütter, and David Lando, 2012, Corporate Bond Liquidity Before and After the Onset of the Subprime Crisis, Journal of Financial Economics 103, 471-92.

Edelen, Roger M, 1999, Investor Flows and the Assessed Performance of Open-End Mutual Funds, Journal of Financial Economics 53, 439–466.

25

Ellul, Andrew, Chotibhak Jotikasthira, and Christian Lundblad, 2011, Regulatory Pressure and Fire Sales in the Corporate Bond Market, Journal of Financial Economics 101, 596-620.

Falato, Antonio, Ali Hortaçsu, Dan Li, and Chaehee Shin, 2016, Fire-Sale Spillovers in Debt Markets, Unpublished working paper.

Feldhütter, Peter, 2012, The same bond at different prices: Identifying search frictions and selling pressures, Review of Financial Studies 25, 1155–1206.

Feroli, Michael, Anil Kashyap, Kermit L. Schoenholtz, and Hyun Song Shin, 2014, Market Tantrums and Monetary Policy, U.S. Monetary Policy Forum Report No.8.

Financial Stability Oversight Committee, 2014 Annual Report.

Goldstein, Itay, Hao Jiang, and David Ng, 2015, Investor Flows and Fragility in Corporate Bond Funds, Unpublished working paper.

Goncalves-Pinto and Breno Schmidt, 2013, Co-Insurance in Mutual Fund Families, Unpublished working paper.

Gorton, Gary and Andrew Metrick, 2010, Regulating the Shadow Banking System, Brookings Papers on Economic Activity.

Gorton, Gary and Andrew Metrick, 2012, Securitized banking and the run on repo, Journal of Financial Economics, Volume 104, Issue 3, Pages 425–451.

Gorton, Gary, and George Penacchi, 1990, Financial Intermediaries and Liquidity Creation, Journal of Finance 45, 49-72.

Greenwood, Robin, and David Thesmar, 2011, Stock Price Fragility, Journal of Financial Economics 102, 471-490.

Hanouna, Paul, Jon Novak, Tim Riley, and Christof Stahel, 2015, Liquidity and Flows of U.S. Mutual Funds, SEC Division of Economic and Risk Analysis White Paper.

Huang, Jiekun, 2013, Dynamic Liquidity Preferences of Mutual Funds, working paper.

Ibert, Markis, Ron Kaniel, Stijn van Nieuwerburgh, and Roine Verstman, 2017, Are Mutual

Fund Managers Paid For Investment Skill?,NBER working paper 23373.

International Monetary Fund, 2015, Global Financial Stability Report: Navigating Monetary Policy Challenges and Managing Risks.

Investment Company Institute, 2016, ICI Comments on the SEC’s Liquidity Risk Management Proposal.

Kacpercyzk, Marcin and Philipp Schnabl, 2010, When safe proved risky, Journal of Economic Perspectives, 24(1), 29-50.

Kacpercyzk, Marcin and Philipp Schnabl, 2013, How safe are money market funds?, Quarterly Journal of Economics 128(3), 1073-1122.

26

Longstaff, Francis A, 2004, The Flight-to-Liquidity Premium in U.S. Treasury Bond Prices, Journal of Business 77, 511-526.

Merrill, Craig B., Taylor D. Nadauld, Shane M. Sherlund, and Rene Stulz, 2012, Did Capital Requirements and Fair Value Accounting Spark Fire Sales in Distressed Mortgage Backed Securities? NBER Working Paper No. 18270.

Moreira, Alan and Alexi Savov, 2016, The Macroeconomics of Shadow Banking, Journal of Finance, forthcoming.

Nagel, Stefan, 2016, The Liquidity Premium of Near-Money Assets, Quarterly Journal of Economics, forthcoming.

Pastor, Lubos, and Robert F. Stambaugh, 2003, Liquidity Risk and Expected Stock Returns, Journal of Political Economy 111, 642-685.

Roll, Richard, 1984, A Simple Implicit Measure of the Effective Bid-Ask Spread in an Efficient Market, The Journal of Finance 39, , 1127-1139.

Schmidt, Lawrence, Allan Timmerman, and Russ Wermers, 2016, Runs on Money Market Mutual Funds, American Economic Review, forthcoming.

Securities and Exchange Commission, 2015. Open-End Fund Liquidity Risk Management Programs, Release No. 33-9922.

Shleifer, Andrei, and Robert W. Vishny, 1992, Liquidation Values and Debt Capacity: A Market Equilibrium Approach, Journal of Finance 47, 1343-1366.

Shleifer, Andrei, and Robert W. Vishny, 1997, The Limits of Arbitrage, The Journal of Finance 52, 35-55.

Simutin, Mikhail, 2014, Cash Holdings and Mutual Fund Performance, Review of Finance 18, 1425-1464.

Sirri, Erik, and Peter Tufano, 1998, Costly Search and Mutual Fund Flows, Journal of Finance 53, 1589-1622.

Stein, Jeremy, 2012, Monetary Policy as Financial-Stability Regulation, Quarterly Journal of Economics 127, 57-95.

Sunderam, Adi, 2015, Money Creation and the Shadow Banking System, Review of Financial Studies 28, 939-977.

Wang, Jeffrey, 2015, Asset Managers and Financial Instability: Evidence of Run Behavior and Run Incentives in Corporate Bond Funds, Harvard College Senior thesis.

Yan, Xuemin (Sterling), 2006, The Determinants and Implications of Mutual Fund Cash Holdings, Financial Management 35, 67-91.

Zeng, Yao, 2015, A Dynamic Theory of Mutual Fund Runs and Liquidity Management, Unpublished working paper.

Table 1Summary Statistics

This table reports summary statistics for the sample of actively managed domestic equitymutual funds studied in the paper. The sample excludes variable annuities, funds closed toinvestors, and funds with less than $100 million in TNA (measured in 2012 dollars). The sampleperiod is January 2003 to June 2016. Illiq is the square-root version of Amihud (2002). ShareOutstanding is the value-weighted average of holdings of each portfolio security relative to thesecurity’s market cap. Manager Overlap is the value-weighted average of the ratio of aggregateholdings of each security s in fund f ’s portfolio by all other funds managed by the same portfoliomanager to the aggregate TNA of all funds managed by fund f ’s portfolio manager. FamilyOverlap is the value-weighted average of the ratio of aggregate holdings of each security s in fundf ’s portfolio by all other funds that belong to the same fund family to the aggregate TNA offunds in the family. Flows is net flows during quarter t scaled by TNA at the beginning of thesix-month reporting period. Fund flows are winsorized at the 1st and 99th percentiles. Turnoveris the minimum of purchases and sales, divided by the monthly average size of the portfolio.σ(Flows) is the standard deviation of monthly net flows during the semi-annual reportingperiod. Sec lending/Shorting/Options/Other are indicators for funds that engage in securitieslending/shorting/trading of options and other derivatives/and other investment practices specifiedin question 70 of form N-SAR.

PercentileN Mean SD 25 50 75

TNA 22535 2046.21 5796.02 231.94 591.15 1573.86Size 22535 6.53 1.32 5.45 6.38 7.36Family size 22535 10.51 2.24 9.09 10.71 12.06Cash/Assets (%) 22535 4.07 4.61 1.15 2.73 5.15∆Cash/Assets (%) 22535 0.43 4.72 −1.14 0.06 1.61∆(Cash/Assets) (%) 22535 −0.16 3.82 −1.29 −0.02 1.19Flowst (%) 22535 2.99 14.12 −3.38 −0.50 4.79Flowst−1 (%) 22535 2.65 12.22 −3.26 −0.16 5.31Illiq (×104) 22535 0.29 0.37 0.10 0.17 0.34σ(Flows) (%) 22535 9.09 10.36 2.43 5.45 11.56Institutional share (%) 22535 31.75 37.64 0.00 11.30 64.07Turnover (%) 22535 53.46 48.61 20.00 39.00 70.00Sec lending 22535 0.47 0.50 0.00 0.00 1.00Shorting 22535 0.02 0.15 0.00 0.00 0.00Options 22535 0.03 0.06 0.00 0.00 0.00Other practices 22535 0.35 0.18 0.25 0.38 0.50Share of Outstanding (%) 22535 0.40 0.70 0.05 0.15 0.42Manager overlap (%) 22535 0.46 0.68 0.01 0.20 0.62Family overlap (%) 22535 0.21 0.32 0.03 0.12 0.27Holdings HHI 22535 0.02 0.04 0.01 0.02 0.03

27

Table 2Flow Management

This table reports results of regressions of changes in cash holdings on monthly fund flows:

∆Cashf,m−6:m = α+5∑

s=0

βs · Flowsf,m−s + εf,m,

where f indexes funds and m indexes months. In columns 1–3, the dependent variable is the changein cash over a six-month period, scaled by assets six months ago. In columns 4–6, the dependentvariable is the change in the cash-to-assets ratio over a six-month period. Independent variables aremonthly net fund flows, scaled by TNA six months ago. Fund flows are winsorized at the 1st and99th percentiles. The sample period is January 2003 to June 2016. Objective-time fixed effects arebased on Lipper objective codes and semi-annual reporting periods. Standard errors are adjustedfor clustering by fund. ∗, ∗∗, and ∗ ∗ ∗ indicate statistical significance at 10%, 5%, and 1%.

∆CashTNA ∆

(CashTNA

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

Flowsf,m 0.167∗∗∗ 0.166∗∗∗ 0.172∗∗∗ 0.068∗∗∗ 0.069∗∗∗ 0.078∗∗∗

(0.011) (0.012) (0.012) (0.008) (0.009) (0.009)Flowsf,m−1 0.100∗∗∗ 0.095∗∗∗ 0.097∗∗∗ 0.020∗∗ 0.010 0.015

(0.013) (0.014) (0.013) (0.009) (0.010) (0.010)Flowsf,m−2 0.086∗∗∗ 0.102∗∗∗ 0.099∗∗∗ 0.021∗∗ 0.031∗∗∗ 0.032∗∗∗

(0.013) (0.014) (0.013) (0.009) (0.010) (0.010)Flowsf,m−3 0.029∗∗ 0.021∗ 0.019 −0.036∗∗∗ −0.043∗∗∗ −0.043∗∗∗

(0.011) (0.013) (0.012) (0.008) (0.009) (0.009)Flowsf,m−4 0.027∗∗ 0.031∗∗ 0.030∗∗ −0.019∗∗ −0.016 −0.021∗∗

(0.011) (0.013) (0.012) (0.009) (0.010) (0.010)Flowsf,m−5 −0.009 −0.016 −0.011 −0.061∗∗∗ −0.071∗∗∗ −0.061∗∗∗

(0.013) (0.015) (0.014) (0.010) (0.011) (0.011)N 22,535 22,535 22,394 22,535 22,535 22,394Adjusted R2 0.144 0.127 0.130 0.030 −0.016 −0.001Objective-time FEs X X X XFund FEs X X X X

28

Table 3Flow Management and Propensity to Internalize Price Impact

This table reports results of regressions of changes in cash holdings on fund flows inter-acted with internalization proxies:

∆Cashf,t = αobj(f),t + β1 · Flowsf,t + β2 · Flowsf,t × Internalizef,t−2 + β3 · Flowsf,t × Illiqf,t−2

+ β4 · Flowsf,t−1 + β5 · Flowsf,t−1 × Internalizef,t−2 + β6 · Flowsf,t−1 × Illiqf,t−2 + εf,t,

where f indexes funds and t indexes quarters. Illiq is the weighted-average across portfoliosecurities of the square-root version of Amihud (2002) for each stock. Share Outstanding is thevalue-weighted average of holdings of each portfolio security relative to the security’s market cap.Manager Overlap is the value-weighted average of the ratio of aggregate holdings of each securitys in fund f ’s portfolio by all other funds managed by the same portfolio manager to the aggregateTNA of all funds managed by fund f ’s portfolio manager. Family Overlap is the value-weightedaverage of the ratio of aggregate holdings of each security s in fund f ’s portfolio by all otherfunds that belong to the same fund family to the aggregate TNA of funds in the family. Allspecifications include objective-time fixed effects. Continuous variables are standardized so thattheir coefficients represent the effect of a one-standard deviation change in each variable. Standarderrors are adjusted for clustering by fund. ∗, ∗∗, and ∗ ∗ ∗ indicate statistical significance at 10%,5%, and 1%.

∆CashTNA ∆

(CashTNA

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

Flowsf,t 0.150∗∗∗ 0.145∗∗∗ 0.145∗∗∗ 0.055∗∗∗ 0.052∗∗∗ 0.052∗∗∗

(0.007) (0.007) (0.007) (0.004) (0.004) (0.004)Flowsf,t−1 0.023∗∗∗ 0.023∗∗∗ 0.022∗∗∗ −0.044∗∗∗ −0.045∗∗∗ −0.046∗∗∗

(0.006) (0.006) (0.006) (0.004) (0.004) (0.004)Flowsf,t × Illiqf,t−2 0.021∗∗∗ 0.030∗∗∗ 0.030∗∗∗ 0.008 0.013∗∗ 0.013∗∗

(0.007) (0.007) (0.007) (0.005) (0.006) (0.006)Flowsf,t−1 × Illiqf,t−2 0.002 0.001 0.002 −0.004 −0.003 −0.003

(0.007) (0.007) (0.007) (0.006) (0.005) (0.005)Illiqf,t−2 0.064 0.057 0.059 0.002 0.000 0.006

(0.048) (0.050) (0.050) (0.038) (0.038) (0.038)Flowsf,t × Share outstandingf,t−2 0.024∗∗ 0.014∗∗∗

(0.010) (0.005)Flowsf,t−1 × Share outstandingf,t−2 −0.000 0.005

(0.011) (0.007)Share outstandingf,t−2 0.031 0.029

(0.029) (0.029)Flowsf,t ×Manager overlapf,t−2 0.017∗∗ 0.009∗∗

(0.008) (0.004)Flowsf,t−1 ×Manager overlapf,t−2 0.002 −0.001

(0.006) (0.004)Manager overlapf,t−2 −0.068∗ −0.043∗∗

(0.036) (0.021)Flowsf,t × Family overlapf,t−2 0.016∗∗ 0.011∗∗

(0.007) (0.004)Flowsf,t−1 × Family overlapf,t−2 0.010 −0.004

(0.007) (0.005)Family overlapf,t−2 0.019 0.021

(0.026) (0.022)N 19,752 19,752 19,752 19,752 19,752 19,752Adjusted R2 0.121 0.120 0.122 0.032 0.031 0.031

29

Tab

le4

Flo

wM

an

agem

ent:

Alt

ern

ati

ve

Exp

lan

ati

on

s

Th

ista

ble

show

sth

ero

bu

stn

ess

of

the

resu

lts

inT

able

3to

alte

rnat

ive

exp

lan

atio

ns

for

the

rela

tion

bet

wee

nin

tern

aliz

atio

np

roxie

san

dfu

nd

’sp

rop

ensi

tyto

acc

omm

od

ate

fun

dfl

ows

thro

ugh

chan

ges

inca

sh.

For

each

regr

essi

onw

ere

por

tth

eβ

2co

effici

ent

onth

ein

tera

ctio

nof

fun

dfl

ows

qu

arte

rt

wit

hth

eb

egin

nin

gof

sem

i-an

nu

alp

erio

dva

lue

ofth

ein

tern

aliz

atio

np

roxy

ind

icat

edby

the

colu

mn

hea

din

g.A

lld

dd

itio

nal

contr

ols

,ex

cep

tfo

rre

turn

s,ar

ein

tera

cted

wit

hqu

arte

rly

fun

dfl

ows.

All

spec

ifica

tion

sin

clu

de

obje

ctiv

e-ti

me

fixed

effec

ts.

Sta

nd

ard

erro

rsare

adju

sted

for

clu

ster

ing

by

fun

d.∗,

∗∗,

and∗∗∗

ind

icat

est

atis

tica

lsi

gnifi

can

ceat

10%

,5%

,an

d1%

.

∆Cash

TNA

∆( Cas

hTNA

)S

har

eM

an

ager

Fam

ily

Sh

are

Man

ager

Fam

ily

outs

tan

din

gov

erla

pov

erla

pouts

tand

ing

over

lap

over

lap

Nβ

R2

βR

2β

R2

βR

2β

R2

βR

2

(1)

Bas

elin

e19

,752

0.02

4∗∗

0.140

0.017∗∗

0.139

0.016∗∗

0.141

0.014∗∗

∗0.

053

0.009∗∗

0.052

0.011∗∗

0.052

(0.0

10)

(0.0

08)

(0.0

07)

(0.0

05)

(0.0

04)

(0.0

04)

Ad

dit

ion

al

contr

ols

:(2

)P

ast

retu

rns

19,6

070.0

18∗

0.142

0.017∗∗

0.142

0.015∗∗

0.143

0.013∗∗

0.054

0.009∗∗

0.053

0.011∗∗

∗0.

053

(0.0

10)

(0.0

07)

(0.0

07)

(0.0

06)

(0.0

04)

(0.0

04)

(3)

Fu

ture

retu

rns

18,2

940.0

19∗∗

0.140

0.015∗∗

0.140

0.017∗∗

0.142

0.014∗∗

0.047

0.010∗∗

0.046

0.012∗∗

0.046

(0.0

09)

(0.0

07)

(0.0

07)

(0.0

06)

(0.0

05)

(0.0

05)

(4)

Pow

ers

ofil

liqu

idit

y19

,752

0.0

22∗∗

0.142

0.018∗∗

0.143

0.017∗∗

∗0.

144

0.014∗∗

0.054

0.009∗∗

0.053

0.011∗∗

0.054

(0.0

10)

(0.0

07)

(0.0

07)

(0.0

05)

(0.0

04)

(0.0

04)

(5)

Dec

iles

ofil

liqu

idit

y19

,752

0.0

23∗∗

0.142

0.018∗∗

0.143

0.017∗∗

0.144

0.014∗∗

∗0.

054

0.008∗

0.053

0.011∗∗

0.053

(0.0

10)

(0.0

08)

(0.0

07)

(0.0

06)

(0.0

04)

(0.0

04)

(6)

Lay

ers

ofli

qu

idit

y19

,752

0.01

6∗

0.150

0.014∗

0.150

0.014∗∗

0.151

0.013∗∗

∗0.

058

0.009∗∗

0.057

0.011∗∗

∗0.

058

(0.0

09)

(0.0

08)

(0.0

07)

(0.0

05)

(0.0

04)

(0.0

04)

(7)

Hol

din

gsH

HI

19,7

520.0

22∗∗

0.141

0.015∗∗

0.141

0.015∗∗

0.142

0.014∗∗

0.053

0.009∗∗

0.052

0.011∗∗

0.053

(0.0

10)

(0.0

07)

(0.0

07)

(0.0

06)

(0.0

04)

(0.0

04)

(8)

Top

shar

e19

,752

0.02

1∗∗

0.141

0.014∗

0.141

0.014∗∗

0.142

0.014∗∗

0.053

0.009∗∗

0.052

0.011∗∗

0.052

(0.0

10)

(0.0

08)

(0.0

07)

(0.0

06)

(0.0

04)

(0.0

04)

(9)

Act

ive

shar

e7,

817

0.0

320.

146

0.028∗∗

0.147

0.047∗∗

∗0.

151

0.019

0.077

0.020∗∗

0.078

0.032∗∗

∗0.

079

(0.0

19)

(0.0

12)

(0.0

11)

(0.0

14)

(0.0

08)

(0.0

09)

(10)

Cli

ente

le19

,752

0.0

25∗∗

0.143

0.017∗∗

0.142

0.015∗∗

0.143

0.015∗∗

∗0.

053

0.009∗∗

0.053

0.011∗∗

0.053

(0.0

10)

(0.0

07)

(0.0

07)

(0.0

05)

(0.0

04)

(0.0

04)

(11)

Fu

nd

man

ager

s18

,949

0.02

3∗∗

0.142

0.018∗∗

0.142

0.016∗∗

0.143

0.013∗∗

0.055

0.008∗

0.054

0.011∗∗

0.054

(0.0

11)

(0.0

08)

(0.0

07)

(0.0

06)

(0.0

04)

(0.0

05)

Sam

ple

spli

ts:

(12)

Sm

all

fund

s9,

778

0.0

58∗

0.175

0.021∗∗

0.175

0.014

0.175

0.027

0.080

0.007

0.079

0.009∗

0.0

79

(0.0

35)

(0.0

10)

(0.0

09)

(0.0

30)

(0.0

06)

(0.0

05)

(13)

Lar

gefu

nd

s9,

974

0.03

5∗∗

∗0.

142

0.008

0.138

0.019∗

0.143

0.017∗∗

0.065

0.009

0.064

0.014∗

0.065

(0.0

11)

(0.0

09)

(0.0

11)

(0.0

07)

(0.0

06)

(0.0

08)

(14)

2003

–200

98,

120

0.01

20.

146

0.027∗∗

0.148

0.010

0.149

0.009

0.050

0.011∗

0.049

0.005

0.050

(0.0

14)

(0.0

12)

(0.0

10)

(0.0

10)

(0.0

07)

(0.0

06)

(15)

2010

–201

611

,632

0.02

5∗∗

0.134

0.009

0.133

0.020∗

0.134

0.015∗∗

∗0.

052

0.007

0.051

0.018∗∗

0.052

(0.0

12)

(0.0

08)

(0.0

10)

(0.0

05)

(0.0

06)

(0.0

08)

30

Tab

le5

Inte

rnalizati

on

an

dFra

gilit

y

Th

ista

ble

rep

ort

sre

sult

sof

regre

ssio

ns

ofst

ock

vol

atil

ity

du

rin

gqu

arte

rt

+1

onin

tern

aliz

atio

np

roxie

s

Vol

s,t+

1=αs

+αt+

1+β·Internalize s

,t+γ′ X

s,t+ε s

,t+

1,

wh

eres

ind

exes

stock

san

dt

ind

exes

qu

arte

rd

ates

.Sto

ckan

dd

ate

fixed

effec

tsar

ein

clu

ded

inal

lsp

ecifi

cati

ons.

Mutualfundsshare

isth

efr

acti

on

of

ou

tsta

nd

ing

own

edby

all

mu

tual

fun

ds.

Fragility

isth

ed

iago

nal

vers

ion

ofG

reen

wood

and

Th

esm

ar(2

011)

stock

frag

ilit

ym

easu

re.

Con

tinu

ous

vari

able

sare

stand

ard

ized

soth

atth

eir

coeffi

cien

tsre

pre

sent

the

effec

tof

aon

e-st

and

ard

dev

iati

onch

ange

inea

chva

riab

le.

Sta

nd

ard

erro

rsar

ead

just

edfo

rcl

ust

erin

gby

stock

.∗,

∗∗,

and∗∗∗

ind

icat

est

atis

tica

lsi

gnifi

can

ceat

10%

,5%

,an

d1%

.

Sh

are

outs

tan

din

gM

an

ager

over

lap

Fam

ily

over

lap

Liq

uid

Illi

qu

idL

iqu

idIl

liqu

idL

iqu

idIl

liqu

id(1

)(2

)(3

)(4

)(5

)(6

)(7

)(8

)(9

)Internalize s

,t-0

.393

∗∗∗

−0.3

40∗∗

−0.4

87∗∗

∗−

0.2

31∗∗

∗−

0.078

−0.3

68∗∗

−0.

182∗

−0.

003

−0.

613∗∗

∗

(0.0

99)

(0.1

72)

(0.1

20)

(0.0

86)

(0.0

99)

(0.1

60)

(0.1

05)

(0.1

15)

(0.2

04)

Fragility s

,t0.

181∗∗

∗0.0

94∗

0.3

51∗∗

∗0.1

81∗∗

∗0.0

95∗

0.3

50∗∗

∗0.1

83∗∗

∗0.

096∗

0.3

47∗∗

∗

(0.0

47)

(0.0

52)

(0.0

97)

(0.0

47)

(0.0

52)

(0.0

97)

(0.0

47)

(0.0

52)

(0.0

97)

Size s

,t-5

.487

∗∗∗

−5.7

56∗∗

∗−

5.8

49∗∗

∗−

5.2

23∗∗

∗−

5.5

91∗∗

∗−

5.5

43∗∗

∗−

5.2

51∗∗

∗−

5.6

26∗∗

∗−

5.5

48∗∗

∗

(0.2

11)

(0.2

96)

(0.3

16)

(0.2

09)

(0.2

98)

(0.3

10)

(0.2

08)

(0.2

97)

(0.3

09)

Mutualfundsshare

s,t

0.48

6∗∗

∗0.2

22

0.636∗∗

∗0.2

67∗

0.0

52

0.273

0.2

45∗

0.0

36

0.248

(0.1

58)

(0.2

09)

(0.2

44)

(0.1

47)

(0.1

88)

(0.2

24)

(0.1

47)

(0.1

92)

(0.2

22)

N13

3,26

767

,043

65,8

44

133,0

87

67,0

38

65,6

69

133,2

37

67,0

43

65,8

14

Ad

just

edR

20.

651

0.6

77

0.619

0.6

51

0.6

77

0.619

0.6

51

0.6

77

0.619

31

Table 6Fund Returns, Flow Pressure, and Internalization

This table reports results of regressions of monthly fund returns on the flow pressure expe-rienced by other funds holding the same securities as fund f :

Rf,m = α+ β0 · Pressuref,m + β1 · Pressure× Internalizef,m + β2 · Internalizef,m−1 + εf,m,

where f indexes funds and m indexes months. Pressuref,m is weighted-average fund flows expe-rienced during month m by other funds holding the same securities as fund f . First, for eachsecurity s held by fund f at the end of the previous quarter, we calculate weighted-average fundflows into all other funds holding security s. The weights are each fund’s holding of security srelative to aggregate holdings of security s by all funds except for fund f . Second, we calculatethe weighted-average across all securities held by fund f , with the weights being portfolio shares.Pressure× Internalizef,m is constructed similarly to Pressuref,m except that each fund’s flow ismultiplied by the fund’s propensity to internalize price impact in a given security. Internalizef,mis the propensity of other funds holding the same securities as fund f to internalize price impactin the securities held by fund f . Internalization measures are standardized in the holdings data.The sample consists of all domestic equity funds in CRSP Mutual Fund Database with at least $10million in TNA and for which the ratio of the value of securities with pressure data to fund TNA isin the [0.75, 1.25] interval. Standard errors are adjusted for clustering by (Lipper) objective-date.∗, ∗∗, and ∗ ∗ ∗ indicate statistical significance at 10%, 5%, and 1%.

Share outstanding Family overlap(1) (2) (3) (4) (5) (6)

Pressuref,t 1.461∗∗∗ 1.642∗∗∗ 1.233∗∗∗ 0.552∗∗∗ 0.641∗∗∗ 0.420∗∗∗

(0.159) (0.174) (0.111) (0.113) (0.128) (0.053)Pressure× Internalizef,t −0.095∗∗ −0.135∗∗∗ −0.121∗∗∗ −0.060∗∗∗ −0.059∗∗∗ −0.039∗∗∗

(0.047) (0.049) (0.026) (0.014) (0.017) (0.010)Internalizef,t 0.001∗∗∗ 0.004∗∗∗ 0.001∗ −0.001∗∗ −0.001∗∗∗ −0.000∗∗

(0.000) (0.001) (0.000) (0.000) (0.000) (0.000)Flowsf,t 0.009∗∗∗ 0.008∗∗∗ 0.002 0.002 0.001 −0.004

(0.002) (0.003) (0.002) (0.003) (0.003) (0.003)N 346,360 346,283 346,260 370,390 370,323 370,357Adjusted R2 0.780 0.782 0.876 0.726 0.728 0.853Date FEs X X X XObjective FEs X XFund FEs X XObjective × Date FEs X X

32

Table 7Cash Holdings

This table reports results of regressions of the cash-to-assets ratio on internalization proxiesand fund characteristics:(

Cash

TNA

)f,t

= α+ β · Internalizef,t + γ′Xf,t + εf,t,

where f indexes funds and t indexes time. All specifications include objective-time fixed effects.Share Outstanding is the value-weighted average of holdings of each portfolio security relative tothe security’s market cap. Manager Overlap is the value-weighted average of the ratio of aggregateholdings of each security s in fund f ’s portfolio by all other funds managed by the same portfoliomanager to the aggregate TNA of all funds managed by fund f ’s portfolio manager. Family Overlapis the value-weighted average of the ratio of aggregate holdings of each security s in fund f ’s portfolioby all other funds that belong to the same fund family to the aggregate TNA of funds in the family.Illiq is the weighted-average across portfolio securities of the square-root version of Amihud (2002)for each stock. Continuous variables are standardized so that their coefficients represent the effectof a one-standard deviation change in each variable. Standard errors are adjusted for clustering byfund. ∗, ∗∗, and ∗ ∗ ∗ indicate statistical significance at 10%, 5%, and 1%.

(1) (2) (3) (4) (5) (6)Share of outstandingf,t 0.737∗∗∗ 0.774∗∗∗

(0.099) (0.100)Manager overlapf,t 0.199∗∗ 0.228∗∗∗

(0.080) (0.083)Family overlapf,t 0.243∗ 0.279∗

(0.136) (0.143)Illiqf,t 0.264∗∗∗ 0.344∗∗ 0.608∗∗∗ 0.657∗∗∗ 0.614∗∗∗ 0.653∗∗∗

(0.087) (0.136) (0.088) (0.134) (0.090) (0.133)σ(Flows)f,t 0.482∗∗∗ 0.513∗∗∗ 0.500∗∗∗ 0.526∗∗∗ 0.501∗∗∗ 0.527∗∗∗

(0.053) (0.054) (0.053) (0.055) (0.053) (0.055)Illiq × σ(Flows)f,t 0.163∗∗∗ 0.164∗∗∗ 0.121∗∗ 0.126∗∗ 0.120∗∗ 0.127∗∗

(0.055) (0.060) (0.056) (0.059) (0.056) (0.059)Sizef,t −0.088 −0.134 0.373∗∗∗ 0.359∗∗∗ 0.321∗∗∗ 0.301∗∗∗

(0.093) (0.096) (0.092) (0.095) (0.089) (0.091)Family sizef,t −0.601∗∗∗ −0.531∗∗∗ −0.699∗∗∗ −0.655∗∗∗ −0.653∗∗∗ −0.605∗∗∗

(0.095) (0.098) (0.102) (0.105) (0.098) (0.101)Institutional sharef,t −0.262∗∗∗ −0.281∗∗∗ −0.342∗∗∗ −0.369∗∗∗ −0.308∗∗∗ −0.333∗∗∗

(0.061) (0.064) (0.063) (0.066) (0.062) (0.065)Turnoverf,t −0.247∗∗∗ −0.255∗∗∗ −0.276∗∗∗ −0.298∗∗∗ −0.279∗∗∗ −0.305∗∗∗

(0.066) (0.066) (0.066) (0.067) (0.066) (0.067)Short sellingf,t 2.942∗∗∗ 2.703∗∗∗ 3.072∗∗∗ 2.873∗∗∗ 3.058∗∗∗ 2.869∗∗∗

(0.717) (0.655) (0.727) (0.670) (0.729) (0.670)Optionsf,t 5.558∗∗∗ 5.304∗∗∗ 5.154∗∗∗ 5.060∗∗∗ 4.952∗∗∗ 4.847∗∗∗

(1.158) (1.189) (1.168) (1.193) (1.149) (1.169)Other practicesf,t −0.231 −0.384 −0.204 −0.348 −0.213 −0.353

(0.360) (0.379) (0.376) (0.393) (0.375) (0.392)Constant 3.900∗∗∗ 3.967∗∗∗ 3.902∗∗∗ 3.960∗∗∗ 3.912∗∗∗ 3.969∗∗∗

(0.148) (0.152) (0.153) (0.157) (0.154) (0.158)N 22,535 22,535 22,535 22,535 22,535 22,535Adjusted R2 0.115 0.125 0.102 0.111 0.103 0.112Time FEs X X XObjective-time FEs X X X

33

Table 8Cash Holdings: Alternative Explanations

This table shows robustness of the results in Table 7 to alternative explanations for the re-lation between cash holdings and internalization proxies. Baseline specifications are from theeven-numbered columns in Table 7, i.e., models that control for objective-date fixed effects. Onlythe coefficient on internalization proxies is reported, other coefficients are omitted for brevity.Standard errors are adjusted for clustering by fund. ∗, ∗∗, and ∗ ∗ ∗ indicate statistical significanceat 10%, 5%, and 1%.

Share outstanding Manager overlap Family overlapN β R2 β R2 β R2

(1) Baseline 22,535 0.774∗∗∗ 0.203 0.228∗∗∗ 0.190 0.279∗ 0.191(0.100) (0.083) (0.143)

Additional controls:(2) Future returns 20,929 0.865∗∗∗ 0.210 0.227∗∗ 0.196 0.282∗∗ 0.197

(0.124) (0.098) (0.144)(3) Past returns 22,355 0.791∗∗∗ 0.207 0.227∗∗∗ 0.194 0.279∗ 0.195

(0.101) (0.083) (0.142)(4) Powers of illiquidity 22,535 0.677∗∗∗ 0.214 0.280∗∗∗ 0.206 0.299∗∗ 0.206

(0.105) (0.079) (0.135)(5) Deciles of illiquidity 22,535 0.681∗∗∗ 0.213 0.279∗∗∗ 0.205 0.294∗∗ 0.205

(0.105) (0.080) (0.134)(6) Layers of liquidity 22,535 0.765∗∗∗ 0.207 0.231∗∗∗ 0.196 0.293∗∗ 0.197

(0.103) (0.084) (0.143)(7) Holdings HHI 22,535 0.733∗∗∗ 0.206 0.172∗∗ 0.194 0.243∗ 0.195

(0.096) (0.085) (0.143)(8) Top share 22,535 0.775∗∗∗ 0.203 0.206∗∗ 0.191 0.259∗ 0.191

(0.100) (0.086) (0.145)(9) Active share 8,889 0.537∗∗∗ 0.233 0.167 0.228 0.505∗∗∗ 0.236

(0.153) (0.126) (0.190)(10) Clientele 22,535 0.764∗∗∗ 0.203 0.227∗∗∗ 0.191 0.272∗ 0.192

(0.100) (0.083) (0.142)(11) Fund managers 21,558 0.785∗∗∗ 0.213 0.183∗∗ 0.199 0.257∗ 0.200

(0.103) (0.084) (0.145)Sample splits:(12) Small funds 11,260 1.084∗∗∗ 0.261 0.252∗∗∗ 0.259 0.221 0.258

(0.353) (0.088) (0.154)(13) Large funds 11,275 0.747∗∗∗ 0.273 0.281∗ 0.256 0.390∗ 0.259

(0.120) (0.152) (0.218)(14) 2003–2009 9,880 0.903∗∗∗ 0.218 0.275∗∗∗ 0.204 0.275∗∗ 0.204

(0.141) (0.090) (0.136)(15) 2010–2016 12,655 0.664∗∗∗ 0.176 0.170 0.165 0.295 0.166

(0.119) (0.106) (0.201)

34

Figure 1Distribution of the Cash/Assets Ratio

This figure shows how the distribution of the Cash/TNA ratio has evolved over time.

0

5

10

15

Cas

h/TN

A

2003h22004h2

2005h22006h2

2007h22008h2

2009h22010h2

2011h22012h2

2013h22014h2

2015h2

10th percentile 25th percentile50th percentile 75th percentile90th percentile

35

Figure 2Distribution of Internalization Measures

This figure shows how the distribution of internalization measures has evolved over time.Manager Overlap is the value-weighted average of the ratio of aggregate holdings of each securitys in fund f ’s portfolio by all other funds managed by the same portfolio manager to the aggregateTNA of all funds managed by fund f ’s portfolio manager. Family Overlap is the value-weightedaverage of the ratio of aggregate holdings of each security s in fund f ’s portfolio by all otherfunds that belong to the same fund family to the aggregate TNA of funds in the family. ShareOutstanding is the value-weighted average of holdings of each portfolio security relative to thesecurity’s market cap.

(a) Manager overlap

0

.005

.01

.015

2003h12004h1

2005h12006h1

2007h12008h1

2009h12010h1

2011h12012h1

2013h12014h1

2015h12016h1

10th percentile 25th percentile50th percentile 75th percentile90th percentile

(b) Family overlap

0

.002

.004

.006

.008

2003h12004h1

2005h12006h1

2007h12008h1

2009h12010h1

2011h12012h1

2013h12014h1

2015h12016h1

10th percentile 25th percentile50th percentile 75th percentile90th percentile

(c) Share outstanding

0

.005

.01

.015

.02

2003h12004h1

2005h12006h1

2007h12008h1

2009h12010h1

2011h12012h1

2013h12014h1

2015h12016h1

10th percentile 25th percentile50th percentile 75th percentile90th percentile

36

Appendix

Table A1Variable definitions

Variable Definition

Active share The percentage of fund holdings that is different from the bench-

mark holdings. Active share is from Martijn Cremers’s website

http://activeshare.nd.edu. Active share is generally reported as of De-

cember of each year. We match observations in N-SAR to the most

recent values of active share assuming the latter is within six months

of the N-SAR reporting date.CashTNA

Cash is cash (74A) + repurchase agreements (74B) + short-term debt

securities other than repurchase agreements (74C) + other investments

(74I) - securities lending collateral. Other investments consist mostly

of money market mutual funds. Value of securities lending collateral is

collected from N-CSR filings. Cash is scaled by total net assets (74T).

Winsorized at the 1st and 99th percentiles.∆CashTNA

The change in cash between two semi-annual reporting periods divided

by TNA as of the previous semi-annual reporting period. Winsorized

at the 1st and 99th percentiles.

∆(CashTNA

)The change in the cash-to-assets ratio between two semi-annual report-

ing periods. Winsorized at the 1st and 99th percentiles.

Clientele In Tables 4 and 8, we control for (a) the number of share classes, (b)

HHI of assets across share classes, and (c) the fraction of assets in share

classes with front load fees.

Experience Number of years managing mutual funds. Using Morningstar data on

the identity of mutual fund managers, we identify when each manager

first started managing mutual funds and measure the number of years

since then. For team-managed funds, Experience is the average across

individual portfolio managers.

Family overlap For each security s held by fund f , we calculate aggregate holdings of

the security by all other funds in the same family and divide this by

family TNA. We then calculate the value-weighted average across all

securities held by fund f . Specifically, Family overlapf =∑

sVf,s∑s Vf,s

×∑fam(j)=fam(f),j 6=f Vf,s∑fam(j)=fam(f) TNAj

, where f and j index funds and s indexes securities.

37

Table A1—Continued

Variable Definition

Family size Log of aggregate TNA across all CRSP mutual funds within the same

family.

Flows Net fund flows during each of the preceding six months (N-SAR item 28)

are scaled by TNA at the end of the previous semi-annual reporting pe-

riod. Net flows are calculated as Total NAV of Shares Sold: New Sales

(Incl. Exchanges) + Total NAV of Shares Sold: Reinv. of Dividends &

Distributions − Total NAV of Shares Redeemed and Repurchased (Incl.

Exchanges). We exclude the Total NAV of Shares Sold: Other as these

capture share activity due to mergers. Flows are winsorized at the 1st

and 99th percentiles.

σ(Flows) Standard deviation of monthly fund flows (28) over the preceding six

months. Fund flows are scaled by TNA as of the beginning of the

semi-annual reporting period.

Fragility Greenwood and Thesmar (2011) measure of fragility. We calculate

the “diagonal” version of fragility that ignores correlation in fund flows

across funds. Column (4) of Table 3 in Greenwood and Thesmar (2011)

shows that the diagonal version of fragility generates similar results

to the full version that accounts for cross-correlation. For each fund

holding security s at time t, we calculate the volatility of fund flows

over the previous five years, requiring each fund to have at least 12

monthly observations. Fund flows are winsorized at the 1st and 99th

percentiles before calculating fund flow volatility.

Fund manager chars In Table 4 and 8, we include controls for (a) team managed funds, (b)

the number of portfolio managers in charge of the fund, (c) mean expe-

rience of the fund’s manager, and (d) CFA credential. Fund manager

data are from Morningstar.

Future returns In Tables 4 and 8, we include controls for average monthly returns over

the subsequent 1, 3, 6, and 12 months.

38

Table A1—Continued

Variable Definition

Illiq We first calculate the square-root version of Amihud (2002) liquidity

measure for each stock in a fund’s portfolio. We use daily data for the

preceding six months. We then calculate the value-weighted average

across all stocks held by a given mutual fund. Portfolio holdings are

from the CRSP Mutual Fund Database. Winsorized at the 1st and

99th percentiles.

Institutional share Fraction of institutional share classes, identified following Chen, Gold-

stein, and Jiang (2010). A share class is considered to be institutional

if a) CRSPs institutional dummy is equal to Y and retail dummy is

equal to N, or b) fund name includes the word institutional or its ab-

breviation, or c) class name includes one of the following suffixes: I, X,

Y, or Z. Share classes with the word retirement in their name or J, K,

and R suffixes are considered to be retail.

Layers of liquidity We sort portfolio securities by their liquidity and measure average liq-

uidity of the top 5%, 10%, 20%, . . . , 90% of the portfolio holdings.

Manager overlap For each security s held by fund f , we calculate aggregate holdings of

the security by all other funds managed by the same portfolio man-

ager and divide this by the aggregate TNA of all funds managed by

the portfolio manager. We then calculate the value-weighted average

across all securities held by fund f . Specifically, Manager overlapf =∑s

Vf,s∑s Vf,s

×∑

mgr(j)=mgr(f),j 6=f Vf,s∑mgr(j)=mgr(f) TNAj

, where f and j index funds and s in-

dexes securities. For funds managed by multiple portfolio managers we

split total holdings of each security equally among the fund’s managers.

Mutual funds share For each stock, we calculate the share of outstanding owned by mutual

funds.

Options Average of 8 binary variables, each equal to one if a fund engages

in writing or investing in 1) options on equities (70B), 2) options on

debt securities (70C), 3) options on stock indices (70D), 4) interest

rate futures (70E), 5) stock index futures (70F), 6) options on futures

(70G), 7) options on stock index futures (70H), and 8) other commodity

futures (70I).

39

Table A1—Continued

Variable Definition

Other practices Average of 7 binary variables for engaging in the following investment

practices: 1) investment in restricted securities (70J), 2) investment in

shares of other investment companies (70K), 3) investment in securi-

ties of foreign issuers (70L), 4) currency exchange transactions (70M),

5) borrowing of money (70O), 6) purchases/sales by certain exempted

affiliated persons (70P), 7) margin purchases (70Q).

Pressure Weighted-average of fund flows experienced by other funds holding the

same securities as fund f :

Pressuref,m =∑s

(wf,s,m−1 ×

(∑j 6=f

Vj,s,m−1∑k 6=f Vk,s,m−1

× Flowsj,mTNAj,m−1

))

where f , j, and k index funds, s indexes securities, and m indexes

(month) dates. Vj,s,m−1 is the dollar holdings of security s by fund j

at time m− 1. wf,s,m−1 is the fraction of fund f ’s portfolio invested in

security s at time m− 1.

Pressure

×InternalizeCalculated similarly to Pressure except that flows into fund j are mul-

tiplied by the fund’s propensity to internalize price impact:

∑s

(wf,s,m−1 ×

(∑j 6=f

Vj,s,m−1∑k 6=f Vk,s,m−1

× Internalizej,s,m−1 ×Flowsj,mTNAj,m−1

))

Because index funds cannot use cash holdings to accommodate fund

flows, their propensity to Internalize is set to zero. When using

Share outstanding as the internalization proxy, the set of holdings is lim-

ited to securities with valid PERMNOs. When using Manager overlap

and Family overlap as internalization proxies, the set of holdings is all

securities with valid CUSIPs.

Redemption fees Binary variable equal to one for funds that impose a deferred or contin-

gent deferred sales load (34) or a redemption fee other than a deferred

or contingent sales load (37).

Sec lending Binary variable equal to one for funds that engage in loaning portfolio

securities (70N).

40

Table A1—Continued

Variable Definition

Share outstanding Weighted-average of the fund’s holdings of each portfolio security rel-

ative to the security’s outstanding amount:∑

sVf,s∑s Vf,s

× Vf,s

Outstandings,

where f indexes funds and s indexes securities.

Short selling Binary variable equal to one for funds that engage in short selling (70R).

Tenure Number of years managing the fund. For team managed funds, Tenure

is the average across individual managers. Manager identities and char-

acteristics are from Morningstar.

Turnover Portfolio turnover for the current semi-annual reporting period (71D).

Portfolio turnover is the minimum of purchases and sales (including

all maturities), divided by the monthly average value of the portfolio.

Portfolio turnover is winsorized at the 1st and 99th percentiles.

41

Table A2Cash Response to Subscriptions versus Redemptions

This table shows the response of cash holdings to subscriptions versus redemptions:

∆Cashf,m = α+5∑

s=0

(β+s Subscriptionsf,m−s + β−s Redemptionsf,m−s

)+ εf,m,

where f indexes funds and m indexes months. In columns 1–3, the dependent variable is the changein cash over a six-month period, scaled by assets six months ago. In columns 4–6, the dependentvariable is the change in the cash-to-assets ratio over a six-month period. Independent variablesare monthly subscriptions and redemptions, scaled by net assets six months ago. Redemptions areexpressed as a nonnegative number. Subscriptions and redemptions are winsorized at the 1st and99th percentiles. The sample period is January 2003 to June 2016. Standard errors are adjustedfor clustering by fund. The table reports the p-values from two-sided tests that the coefficients onsubscriptions and redemptions are equal in magnitude and opposite in sign. ∗, ∗∗, and ∗∗∗ indicatestatistical significance at 10%, 5%, and 1%.

∆CashTNA ∆

(CashTNA

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

Subscriptionsf,m 0.157∗∗∗ 0.160∗∗∗ 0.165∗∗∗ 0.065∗∗∗ 0.064∗∗∗ 0.072∗∗∗

(0.014) (0.015) (0.014) (0.008) (0.010) (0.009)Redemptionsf,m −0.116∗∗∗ −0.113∗∗∗ −0.113∗∗∗ −0.079∗∗∗ −0.093∗∗∗ −0.092∗∗∗

(0.025) (0.024) (0.024) (0.017) (0.017) (0.018)Subscriptionsf,m−1 0.094∗∗∗ 0.088∗∗∗ 0.089∗∗∗ 0.011 −0.001 0.004

(0.013) (0.013) (0.013) (0.011) (0.011) (0.011)Redemptionsf,m−1 −0.073∗∗∗ −0.059∗∗∗ −0.067∗∗∗ −0.032∗∗ −0.033∗∗ −0.036∗∗

(0.016) (0.017) (0.016) (0.015) (0.015) (0.014)Subscriptionsf,m−2 0.092∗∗∗ 0.106∗∗∗ 0.104∗∗∗ 0.027∗∗ 0.036∗∗∗ 0.037∗∗∗

(0.014) (0.015) (0.014) (0.010) (0.013) (0.012)Redemptionsf,m−2 −0.069∗∗∗ −0.067∗∗∗ −0.065∗∗∗ −0.025 −0.036∗ −0.033∗

(0.020) (0.024) (0.023) (0.017) (0.018) (0.018)Subscriptionsf,m−3 0.022∗ 0.015 0.014 −0.030∗∗ −0.036∗∗ −0.038∗∗∗

(0.011) (0.012) (0.012) (0.013) (0.013) (0.013)Redemptionsf,m−3 −0.001 −0.001 −0.001 0.033∗∗ 0.021 0.023

(0.014) (0.016) (0.015) (0.013) (0.014) (0.014)Subscriptionsf,m−4 0.005 0.011 0.007 −0.030∗∗∗ −0.025∗∗ −0.032∗∗∗

(0.013) (0.013) (0.013) (0.009) (0.011) (0.010)Redemptionsf,m−4 −0.066∗∗∗ −0.049∗∗∗ −0.047∗∗∗ −0.015 −0.017 −0.020

(0.016) (0.017) (0.015) (0.015) (0.013) (0.012)Subscriptionsf,m−5 −0.008 −0.014 −0.007 −0.051∗∗∗ −0.063∗∗∗ −0.051∗∗∗

(0.013) (0.015) (0.014) (0.009) (0.011) (0.011)Redemptionsf,m−5 −0.034 −0.024 −0.021 0.027 0.014 0.012

(0.021) (0.023) (0.022) (0.019) (0.018) (0.020)N 22,535 22,535 22,394 22,535 22,535 22,394Adjusted R2 0.145 0.128 0.131 0.031 −0.014 0.000p0 0.167 0.129 0.083 0.449 0.128 0.301p1 0.218 0.128 0.214 0.236 0.084 0.077p2 0.363 0.208 0.200 0.939 0.979 0.884p3 0.218 0.467 0.514 0.827 0.407 0.421p4 0.003 0.079 0.015 0.012 0.024 0.001p5 0.150 0.217 0.347 0.303 0.019 0.104poverall 0.021 0.160 0.085 0.008 0.000 0.000Objective-time FEs X X X XFund FEs X X X X 42

(a) Fund size

0

.005

.01

.015

Shar

e ou

tsta

ndin

g

4 5 6 7 8 9 10Fund size

(b) Family size

.002

.004

.006

.008

Shar

e ou

tsta

ndin

g

6 7 8 9 10 11 12 13 14Family size

(c) Illiquidity

.002

.004

.006

.008

.01

Shar

e ou

tsta

ndin

g

-.00002 0 .00002 .00004 .00006 .00008 .0001Illiquidity

(d) Flow volatility

.0035

.004

.0045

.005

Shar

e ou

tsta

ndin

g

0 5 10 15 20 25 30 35 40Flow volatility

(e) Turnover

.002

.003

.004

.005

.006

Shar

e ou

tsta

ndin

g

0 50 100 150 200Turnover

(f) Institutional share

.002

.003

.004

.005

.006

Shar

e ou

tsta

ndin

g

-20 0 20 40 60 80 100Institutional share

(g) Experience

.003

.004

.005

.006

.007

Shar

e ou

tsta

ndin

g

0 5 10 15 20 25Experience

(h) Tenure with fund family

.002

.003

.004

.005

.006

.007

Shar

e ou

tsta

ndin

g

0 5 10 15 20 25Tenure with fund family

(i) Holdings HHI

.002

.004

.006

.008

Shar

e ou

tsta

ndin

g

0 .02 .04 .06 .08 .1Holdings HHI

Figure A1. Share outstanding and fund characteristics. This figure shows binned scatterplots of share outstandingversus various fund characteristics. Observations are sorted into 20 equal-sized bins based on the distribution of a givenfund characteristic. Each dot represents average values in a given bin. All panels adjust for objective-time fixed effects,and all panels except for (a) control for fund size.

43

(a) Fund size

.002

.003

.004

.005

.006

Man

ager

ove

rlap

4 5 6 7 8 9 10Fund size

(b) Family size

.001

.002

.003

.004

.005

.006

Man

ager

ove

rlap

6 7 8 9 10 11 12 13 14Family size

(c) Illiquidity

.002

.003

.004

.005

.006

Man

ager

ove

rlap

-.00002 0 .00002 .00004 .00006 .00008 .0001Illiquidity

(d) Flow volatility

.004

.0045

.005

.0055

Man

ager

ove

rlap

0 5 10 15 20 25 30 35 40Flow volatility

(e) Turnover

.0035

.004

.0045

.005

.0055

Man

ager

ove

rlap

0 50 100 150 200Turnover

(f) Institutional share

.004

.0045

.005

.0055

.006

Man

ager

ove

rlap

-20 0 20 40 60 80 100Institutional share

(g) Experience

.002

.003

.004

.005

.006

.007

Man

ager

ove

rlap

0 5 10 15 20 25Experience

(h) Tenure with fund family

.003

.0035

.004

.0045

.005

.0055

Man

ager

ove

rlap

0 5 10 15 20 25Tenure with fund family

(i) Holdings HHI

.002

.004

.006

.008

Man

ager

ove

rlap

0 .02 .04 .06 .08 .1Holdings HHI

Figure A2. Manager overlap and fund characteristics. This figure shows binned scatterplots of manager overlapversus various fund characteristics. Observations are sorted into 20 equal-sized bins based on the distribution of a givenfund characteristic. Each dot represents average values in a given bin. All panels adjust for objective-time fixed effects,and all panels except for (a) control for fund size.

44

(a) Fund size

.0016

.0018

.002

.0022

.0024

.0026

Fam

ily o

verla

p

4 5 6 7 8 9 10Fund size

(b) Family size

.001

.0015

.002

.0025

.003

Fam

ily o

verla

p

6 7 8 9 10 11 12 13 14Family size

(c) Illiquidity

.0018

.002

.0022

.0024

.0026

.0028

Fam

ily o

verla

p

-.00002 0 .00002 .00004 .00006 .00008 .0001Illiquidity

(d) Flow volatility

.0019

.002

.0021

.0022

.0023

.0024

Fam

ily o

verla

p

0 5 10 15 20 25 30 35 40Flow volatility

(e) Turnover

.0018

.002

.0022

.0024

.0026

.0028

Fam

ily o

verla

p

0 50 100 150 200Turnover

(f) Institutional share

.0015

.002

.0025

.003

Fam

ily o

verla

p

-20 0 20 40 60 80 100Institutional share

(g) Experience

.0015

.002

.0025

.003

.0035

Fam

ily o

verla

p

0 5 10 15 20 25Experience

(h) Tenure with fund family

.0015

.002

.0025

.003

.0035

.004

Fam

ily o

verla

p

0 5 10 15 20 25Tenure with fund family

(i) Holdings HHI

.001

.002

.003

.004

.005

Fam

ily o

verla

p

0 .02 .04 .06 .08 .1Holdings HHI

Figure A3. Family overlap and fund characteristics. This figure shows binned scatterplots of family overlap versusvarious fund characteristics. Observations are sorted into 20 equal-sized bins based on the distribution of a given fundcharacteristic. Each dot represents average values in a given bin. All panels adjust for objective-time fixed effects, andall panels except for (a) control for fund size.

45