Strategic Complementarities and Mutual Fund Runsbschool.nus.edu/Portals/0/images/CAMRI/Research Papers/fundRuns.pdf · Strategic Complementarities and Mutual Fund ... runs in mutual

Strategic Complementarities and Mutual Fund Runs

Meijun Qian* A. Başak Tanyeri**

National University of Singapore Bilkent University

February 2010

Abstract

Are self-fulfilling runs possible in mutual funds? This paper provides insights by investigating whether anticipation of adverse events can trigger runs in mutual funds. The adverse event in question is the litigations concerning market-timing and late trading practices that were filed in 2003 and 2004. We find that pre-event runs start as early as six months before litigation announcements. The size of pre-event runs is about half the size of the runs after litigation announcements. Investors, who run before litigation announcements, earn significantly higher risk- and peer- adjusted returns than do those who run after, especially in funds holding illiquid assets and in funds incurring large outflows. The return difference is driven by the fire-sale costs because event returns of firms held by implicated funds with negative flows are significantly negative. Our analysis suggests that pro-rata-ownership design may not suffice to prevent runs in the mutual funds. Return differences due to the timing of withdrawals suggest strategic complementarities in the fund industry, where investors have incentives to withdraw in anticipation of other investors doing so.

JEL: G23 G14

Keywords: Runs, mutual fund flows, returns, and strategic complementarities. --------------------------------------------------------------------------------------------------------------------------------- Meijun Qian is an Assistant Professor of Finance at NUS Business School. Tel: (65) 6516 8119; e-mail: [email protected] Basak Tanyeri is an Assistant Professor of Finance at Bilkent University. Tel: (90) 312-290-1871; e-mail: [email protected]

  2

Strategic Complementarities and Mutual Fund Runs

Abstract

Are self-fulfilling runs possible in mutual funds? This paper provides insights by investigating whether anticipation of adverse events can trigger runs in mutual funds. The adverse event in question is the litigations concerning market-timing and late trading practices that were filed in 2003 and 2004. We find that pre-event runs start as early as six months before litigation announcements. The size of pre-event runs is about half the size of the runs after litigation announcements. Investors, who run before litigation announcements, earn significantly higher risk- and peer- adjusted returns than do those who run after, especially in funds holding illiquid assets and in funds incurring large outflows. The return difference is driven by the fire-sale costs because event returns of firms held by implicated funds with negative flows are significantly negative. Our analysis suggests that pro-rata-ownership design may not suffice to prevent runs in the mutual funds. Return differences due to the timing of withdrawals suggest strategic complementarities in the fund industry, where investors have incentives to withdraw in anticipation of other investors doing so.

JEL: G23 G14

Keywords: Runs, mutual fund flows, returns, and strategic complementarities.

  3

1. Introduction

The first-come-first-served principle governing deposit withdrawals motivates

bank runs. Every depositor wants to withdraw before others do because those at the back

of the line may not recover their deposits (Diamond and Dybvig, 1983; Chari and

Jagannathan, 1988). In contrast, mutual funds allocate the proceeds from asset sales on a

pro-rata basis. The pro-rata design should shield mutual funds from runs. However,

mutual funds may prove susceptible to runs if the timing of redemptions affects the

returns to shareholders. Such a situation can happen when shareholders who redeem early

pass the cost of asset sales on to shareholders who redeem late1. If returns are higher for

early withdrawals than for late withdrawals, investors will have incentives to exit before

others do. This setting gives rise to strategic complementarities, where investors’

incentive to take an action increases if they believe more investors will take the same

action (Bulow et. al., 1985; Chen et. al., 2009). This paper aims to provide direct

evidence of such strategic complementarities in the mutual fund industry and an

economic rationale for fund runs.

We define a a pre-event run as concerted redemption of mutual fund shares in

anticipation of an adverse event and run as concerted redemption upon revelation of such

an event. The adverse event that this paper focuses on is the litigations of 2003 and 2004

alleging that certain mutual funds allowed some investors to engage in late trading and

1Mutual funds face two types of transaction costs when selling portfolio assets: trading costs and costs arising from dilution effects of flows (Edelen, 1996). Edelen (1999) shows that liquidity-motivated trading hurts fund performance. Chen, Goldstein, and Jiang (2009) argue that the costs of liquidity-motivated trading are partially borne by existing shareholders and may cause strategic (or payoff) complementarities in mutual fund shares.

  4

market timing2. Investors who engage in late trading and market timing enjoy profits at

the expense of investors who do not engage in these practices. Upon the suspicion or the

revelation that fund managers do not serve the interests of all investors equally, investors

that are at a disadvantage may discipline implicated funds by withdrawing existing

investments and/or by withholding new investments.

We investigate whether there are pre-event runs and runs after litigation

announcements. We focus on pre-event runs to examine whether there are incentives in

addition to market discipline that might explain the redemptions of and/or lack of new

investments in implicated funds. If the timing of runs matters, e.g., it is beneficial to

withdraw before others; investors will run funds as long as they fear other investors will

do so regardless of whether there is revelation of adverse information. Consequently,

mutual fund industry may be exposed to financial fragility generated by strategic

complementarities.

We examine the return differences for withdrawals at different times to show

whether rationale of strategic complementarities exists. The concerted redemption and

the lack of new sales that follow litigation announcements would force funds to liquidate

assets quickly. Coval and Stafford (2007) find that large selling-volume by institutional

investors temporarily depresses underlying asset prices. Shareholders who redeem shares

at this time will suffer losses. Investors who can anticipate litigations and the redemptions

that would follow, have incentives to redeem shares before litigation announcements. By

2 Late trading is the purchase or sale of mutual fund shares after four p.m. at Net Asset Value (NAV) determined at four p.m. Market timing is the short-term trading of mutual fund shares to exploit price inefficiencies between mutual fund shares and underlying securities in the funds’ portfolios. Bank of America Nations Fund Securities Litigation Complaint is a representative case describing in detail the allegations of market timing and late trading (http://securities.stanford.edu/1028/BAC03-01/20030905_f01c_Lin.htm ).

  5

exiting early, informed investors avoid the fire-sale costs caused by concerted future

withdrawals.

Furthermore, the return differences due to the timing of withdrawals will be larger

if the funds hold more illiquid assets or if the funds are likely to suffer from larger

outflows. Therefore, the incentive of early runs would also be greater for these funds.

Our paper supports the above arguments by providing empirical evidence for

these four questions. First, are there pre-event runs and runs post litigation

announcements? Second, do investors who run prior to announcements enjoy financial

benefits compared to investors who run post? Third, is the return difference larger in

funds with less liquid assets or large outflows? Finally, are the low returns to late

withdrawals caused by fire-sale costs?

We find fund runs both prior to and post litigation announcements. Pre-event runs

start as early as six months before litigation announcements. Flows to implicated funds

prove 2.28% lower than flows to non-implicated funds in the six months before litigation

announcements and are continuously lower for at least two years after. Second, investors

who run before litigation announcements earn significantly higher risk- and peer-

adjusted returns than do those who run after litigations. The difference in returns is as

high as 2.17% accumulated from six months before announcements to six months after.

Third, not all funds prove equal in their vulnerability to runs. Funds holding illiquid

assets and funds that are expected to suffer from large outflows, such as disreputable

funds (as measured by the history of Security Exchange Commission (SEC)

investigations) experience more severe runs both prior to and post litigation

announcements. Moreover, the return difference between investors who run before

  6

litigation announcements and those who run after proves more pronounced in funds that

hold illiquid assets, funds that suffer from larger outflows, and funds with no prior history

of SEC charges. Finally, we show that the return difference between early and late

redemptions is driven by fire-sale costs. Event returns of firms held by implicated funds

with negative flows are negative and significant.

Our results indicate that mutual fund investors who anticipate negative flows

motivated by adverse events such as litigations have incentives to withdraw early and

avoid fire-sale costs. The incentives for early exit have an important implication in that

investors may run funds in expectation of other investors doing so. When the timing of

the action (runs) matters for payoff (returns), strategic complementarities exist. Strategic

complementarities can amplify the impact of adverse events on fundamentals and

generate financial fragility. However, we may not observe mutual fund runs to the extent

of bank runs unless there is a systematic liquidity shock to all fund investors (Chen,

Goldstein, and Jiang 2008). In the absence of such a liquidity shock, other investors will

purchase the assets in fire-sale and might correct the mispricing (Hanson, Hong, and

Stein, 2008)3.

The decision of the US Treasury to insure the holdings of eligible money-market

mutual funds in the wake of the turmoil caused by the run on the money-market mutual

fund Reserve Primary Fund in September 2008 showcases the financial fragility that the

mutual fund industry faces4. Our findings explain why runs can happen in mutual funds

3 Chen, Hanson, Hong, and Stein (2008) show that hedge funds that purchase funds’ underlying assets at the depressed price during fire-sale periods generate arbitrage profits similar to the profits of the short sellers. However, short selling is not allowed in most mutual funds. 4 The Reserve Primary Fund held debt securities of Lehman Brothers; following the bankruptcy of Lehman Brothers, redemptions totaled about two-thirds of Total Net Assets (Wall Street Journal, 2008; New York

  7

and the events surrounding the demise of Lehman Brothers underline the financial

fragility of the mutual fund industry.

This paper is first to investigate and document runs in mutual fund industry in

anticipation of litigation announcements. Furthermore, we provide an economic rationale

-- to avoid fire-sale costs-- for why investors want to run before others do. These findings

contribute to our understanding of how liquidity associated trading costs may generate

strategic complementarities and financial fragility in the mutual fund industry.

The rest of the paper proceeds as follows: Section 2 develops the methodology.

Section 3 describes the data. Section 4 discusses empirical results. Section 5 concludes

the paper.

2. Research Methods

Empirically, we address four questions. The first question investigates whether

informed investors run implicated funds prior to litigation announcements. The second

question examines whether investors who run funds prior to the filing of lawsuits realize

higher returns than do those who run post. The third question analyzes whether some

types of funds are more susceptible to silent runs. The last question investigates whether

the low returns on withdrawals after litigation are caused by fire-sale costs of underlying

assets.

2.1. Detecting pre-event runs

Times, 2008). The liquidity crunch in the short-term credit market meant that non-redeeming investors would bear the fire-sale costs associated with asset sales to satisfy redemptions. The Treasury stated its concerns about the ensuing uncertainty in the mutual fund industry and in instating the guarantee program as follows: “…Maintaining confidence in the money market fund industry is critical to protecting the integrity and stability of the global financial system. …This action should enhance market confidence and alleviate investors' concerns about the ability for money market mutual funds to absorb a loss…” (US Treasury Department Press Release, 19 September 2008).

  8

We need benchmarks of “normal” flow to document pre-event runs. The first

benchmark is flows to peers who are not named in the 2003 and 2004 lawsuits. We

construct three groups of funds. The first group consists of funds whose management

companies are not involved in the litigations of 2003 and 2004 (named as funds in non-

implicated families). The second group includes funds that are not named in the suits but

whose management companies are (non-implicated funds in implicated families). The

third group consists of the funds named in the suits (implicated funds).

We compute fund flows as follows:

Flowi,t = [TNAi,t –TNAi,t-1 *(1+ri,t)] / TNAi,t-1 , (1)

where Flowi,t is net flows of fund i in month t. TNAi,t-1 and TNAi,t are total net assets of

fund i in month t-1and t, respectively. ri,t is return of fund i in month t. We compare net

flows of the three groups around the litigation dates to detect whether implicated funds

have lower flows than do non-implicated funds.

The second benchmark for ‘normal’ flows is estimated net flows from a model

that tries to capture main determinants of fund flows. We review the empirical literature

to develop the flow model. Gruber (1996), Chevalier and Ellison (1997), Sirri and Tafuno

(1998), Zheng (2000), and Del Guercio and Tkac (2001, 2002) all show that past returns

predict future flows. Qian (2008) finds that industry-level and style-level flows explain

individual fund-level flows. We use a model of flows that includes variables for fund

characteristics, past returns, fund-level, and style-level flows. To detect pre-litigation and

post-litigation runs, we construct 25 event-window dummies.

The resulting model of flows is:

Flowi,t = a + ∑bj * fund characteristicsi,t j + ∑cj * past returnsi

j

  9

+ ∑dj * aggregate flowst j + ∑γj * Event-window Dummiesi,t

j +εi,t, (2)

where fund characteristics include size, age, 12b-1 fees, rear, and front loads. Size is log

of TNA. Age is log of the days since the first offer date. 12b-1 fees are the annual fees

paid to financial advisors, measured as percentage of TNA. Front and rear loads are the

charges for purchases and redemption of shares measured as percentage of TNA. Past

returns include compounded returns in the past one (Ri,t-1), three ((1+ Ri,t-1)* (1+ Ri,t-2)*

(1+ Ri,t-3)-1), and six months ((1+ Ri,t-1)* (1+ Ri,t-2)* (1+ Ri,t-3)* (1+ Ri,t-4)* (1+ Ri,t-5)* (1+

Ri,t-6)-1). Aggregate flows include industry- and style-level flows. Industry-level flows are

the sum of flows in dollars (Σi (TNAi,t –TNAi,t-1*(1+Ri,t ))) to all funds in the sample

divided by the sum of lagged TNA (Σi (TNAi,t-1)). Style-level flows are the sum of flows

in dollars to all funds with the same investment style divided by the sum of lagged TNA.

We adopt the style classification of Pastor and Stambaugh (2002) and Ferson and Qian

(2005). There are eight styles: aggressive growth, growth-income, global equity, other

equity, bond funds, municipal funds, money market, and other. Event-window dummy (n

month) equals 1 if it is the nth month from the date of the litigation filing, and 0 otherwise

(n = -1, -2…12, 0, 1, 2…12).

2.2 Rationale for pre-event runs

What incentives exist for shareholders to run a mutual fund when proceeds from

asset sales are determined by the prices of underlying assets and are distributed pro-rata?

Investors may see abusive behavior as indicative of how faithfully fund managers serve

their best interests. As such, investors may want to redeem shares as soon as they are

informed, privately or publicly, of abusive practices such as market timing or late trading,

in the funds they invest in. If and when sufficient numbers of investors learn of abusive

  10

behavior in a fund, a run may ensue. Mutual funds must liquidate assets quickly in order

to satisfy the redemption of shares. If large selling volumes temporarily depress

underlying asset prices, shareholders who redeem shares at this point of time would

realize negative abnormal returns.

We examine whether there are financial benefits to redeeming shares prior to the

revelation of abusive behavior. We develop two approaches to detect the return

differences between investors who run funds prior to lawsuits being filed and investors

who run post. The first approach benchmarks ‘normal’ returns using five different models

for returns and introduces indicators for pre- and post-event months to identify return

differences. The return models are the market model (Sharpe, 1964; Lintner, 1965), the

market model with lagged market returns (Scholes and William, 1977), the Fama-French

benchmarks (Fama and French, 1992 and 1993), the Fama-French benchmark with a

fourth factor that captures momentum (Jegadeesh and Titman, 1993; Carhart, 1997), and

the market model with a factor that captures liquidity (Pastor and Stambaugh, 2003). The

five models of returns are:

ri,t = α + β*r m.t + ∑αn * Dummyn+ ε i,t, (3)

ri,t = α + β1*r m.t + β2*rm.t-1 + ∑αn * Dummytn + ε i,t , (4)

ri,t = α + ∑βj * FFtj + ∑αn * Dummyt

n+ ε i,t, (5)

ri,t = α + ∑βj * FFtj + γ1* MOMt + ∑αn * Dummyt

n + ε i,t, (6)

ri,t = α + β*rm.t + γ2*LIQt+ ∑αn * Dummytn + ε i,t . (7)

where ri,t is the excess returns (net of the risk-free rate) of fund i on month t. rm.t is the

excess market return on month t. FFj include market returns, size (SMB) and value

(HML) factors. MOM is the momentum factor and LIQ is the liquidity factor. Event-

  11

window dummy (n month) equals 1 if it is the nth month from the date of the litigation

filing, and 0 otherwise (n = -1, -2…-6, 0, 1, 2…6).

2.3. Impact of fund characteristics and liquidity on pre-event runs

Pre-event runs are motivated by suspicions of litigations and liquidation costs that

would arise to satisfy the redemptions following litigations. Factors that influence

investors’ belief or awareness of abusive behavior and factors that would increase fire-

sale costs would affect investors’ decision to run before the adverse event is confirmed.

Fund and investor characteristics such as management reputation and the ability of

investors to collect and process information may affect whether and when investors

become aware of abusive behavior. Ownership structure and a history of SEC charges

measure fund reputation. Investors may judge funds in conglomerate families to be less

likely to engage in abusive behavior since loss of reputation would hurt the abused fund

as well as other businesses of the conglomerate. Hence, the consequences of abusive

behavior may be larger for conglomerates than for fund families that only focus on

managing mutual funds. Past actions may predict future decisions. As such, investors

may judge funds with no history of abusive behavior to be less likely to engage in

abusive behavior in the future. We use differences in the distribution channels for fund

shares to measure the information collection and processing ability of investors. Investors

aided by financial advisors may be in a better position to judge which funds are more

likely to engage in abusive behavior. We expect investors who are assisted by financial

advisors to be more likely to anticipate abusive behavior and redeem shares in implicated

funds prior to lawsuits being filed.

  12

To investigate whether fund and investor characteristics would influence the

susceptibility of funds to silent runs, we generate dummy variables for these

characteristics. Conglomerate, charge history, and 12b-1 fees indicators take on the value

1 if the fund is part of a conglomerate, has had a SEC investigation in the past eight

years, and charges 12b-1 fees, respectively and 0 otherwise. Retail fund and institutional

fund indicators take on the value 1 if the fund is a retail or institutional fund, respectively,

and 0 otherwise. We interact the dummy variables for fund and investor characteristics

with all the variables in Equation (2).

The economic rationale for silent runs is the liquidation cost (price depression)

that funds bear when they are forced to sell assets upon the revelation of an adverse

event. The liquidity costs increase with the illiquidity of underlying assets and with the

volume of redemptions. Investors in funds with illiquid assets, such as Real Estate

Investment Trusts (REITs), international assets, or municipal funds have stronger

incentives to run since benefits to running may be greater.

To investigate the impact of the liquidity of underlying assets on run incentives

and the benefits that investors realize from running early, we generate a dummy variable

(liquid) that identifies liquid funds. We classify funds as liquid and illiquid based on the

assets they invest in, as defined in the style classification. Liquid funds invest in large-cap

stock and Treasury bills. Illiquid funds invest in small-cap stocks, sector stocks,

international equity and bonds, and asset-backed securities. We interact the liquid dummy

with all the variables in Equations (3)-(7).

We conduct these analyses using a two-step fund-by-fund approach as well as

using a panel approach. The panel approach is efficient in the sense that it pools

  13

information of all funds, however, may suffer from the problem that coefficients are

forced to be the same for all funds. In the fund-by-fund estimation, the first step estimates

the flow models and return models for each fund using time series observations only. The

five return models are the market model (Sharpe, 1964; Lintner, 1965), the market model

with lagged market returns (Scholes and William, 1977), the Fama-French benchmarks

(Fama and French, 1992 and 1993), the Carhart Four factor models that captures

momentum (Jegadeesh and Titman, 1993; Carhart, 1997), and the market model with a

factor that captures liquidity (Pastor and Stambaugh, 2003). The control variables for

flow analysis include accumulated returns in the past six, three, and one month, industry

level flows, and style level flows. In addition, the explanatory variables include two

indicators for six months pre- and post-event respectively. The coefficients on these

indicators estimate the silent runs and runs (from the flow-model estimation) and risk-

adjust returns (from the return-model estimation) six months pre- and post- event. The

second step compares the estimated silent runs, and risk-adjusted returns in the cross

section. We investigate their differences by groups of funds. The funds are grouped

according to their SEC charge history, ownership structure, distribution channels (proxied

by 12-b1 fees), investor clienteles, the liquidity of underlying assets, and the magnitude

of outflows in the post-event window.

2.4 Cost of fire-sales

A direct test of whether mutual funds bear costs associated with liquidating

portfolio positions involves analyzing the returns to underlying assets of fund portfolios.

To this end, we first identify implicated funds that face negative daily flows around

litigation announcements and further categorize them by whether the funds have negative

  14

flows in every week of September 2003. Second, we compile the holdings of these funds

and catalog which shares they choose to sell and which shares to purchase. We measure

the changes of shares held by fund portfolio from the nearest quarterly statement prior to

September 2003 to the nearest statement post. Third, we calculate abnormal returns to

sold shares in the event windows surrounding the sale. Since the precise dates for when

funds trade is not available due to the quarterly reporting of fund holdings, we pick

September 2003 as the month in which the trades of implicated funds would most reflect

the withdrawals associated with the litigations.

To compute cumulative abnormal returns (CARs), we estimate the market model

for each firm using daily returns from 282 days to 30 days prior to litigation

announcements. The market model uses CRSP equally-weighted-portfolio as the market

portfolio. Weekly CARs are then aggregated using estimated daily abnormal returns for

the weeks of September 2-4, September 7-11, September 14-18, September 21-25, 2003.

3. Data

To identify funds and fund families implicated in the litigations concerning

market timing and late trading, we conduct a keyword search in the Financial Times5 and

the Wall Street Journal. We also search SEC litigation filings of Stanford Law School

Securities Class Action Clearinghouse6. Table 1 summarizes the results of the search

process. The table lists names of implicated fund-families, activities that they are indicted

5We use three key words --- investigation, mutual fund, and Spitzer--- to search the Financial Times and Wall Street Journal between September 3, 2003 and December 31, 2005. 6Stanford Law School Securities Class Action Clearinghouse (available online at http://securities.stanford.edu/index.html) compiles detailed information relating to the prosecution, defense, and settlement of federal class-action securities fraud litigations.

  15

for, regulatory authorities involved, litigation announcement dates, and names of the

parent companies.

New York Attorney General Eliot Spitzer filed a complaint in the New York

Supreme Court on September 3, 2003, alleging that the mutual fund companies of Bank

of America Corp., Bank One Corp., Janus Capital Group Inc., and Strong Capital

Management Inc., allowed certain hedge fund managers to trade illegally in their fund

units. Mr. Spitzer’s complaint marked the beginning of a formal investigation into the

pricing practices of mutual fund companies. From September 2003 to August 2004,

SEC, the New York State Attorney General, and other regulatory authorities filed

litigations concerning funds in 25 mutual fund families.

Another two dates are also important to shred lights on how belief of adverse

events triggers silent runs and runs: the first date when information from trustable sources

triggers investors suspicion of abusive behavior and potential investigation in mutual

funds and the first date when information from trustable sources indicates the actual

investigation and possible SEC charges into the timing and late-trading activities in

funds. In fact, SEC was aware of the fair pricing problems in mutual fund as far back as

in 1997 and the probe of hedge fund trades that take advantage of such problems was

under the way since 2002. However, there was little suspicion on the active cooperation

from mutual fund management side. We identify a news article through Lexis-Nexus on

March 5, 2003 that indicated the possible active involvement of mutual fund

management. Moreover, by March 26, 2003 the pressure from congress to strengthen

mutual fund regulation peaks. Therefore, march 2006 is a reasonable time when investors

start to suspect abusive behavior and potential investigation in mutual funds. Although,

  16

September 3rd , 2003 is the date when Spitzer filed the formal complaint. A wall street

journal article on September 1st, 2003 has already revealed the investigation going on and

formal complaints any time. Therefore, September 1st is when the public announcement is

actually made though relatively softly.

We rely on the CRSP mutual funds survival-bias-free database for the universe of

mutual funds (database provided by WRDS). The database provides monthly

observations of funds’ total net assets (TNA) and returns (R). We merge our list of

implicated funds with the CRSP universe of funds using ticker symbols. The resulting

sample identifies funds as implicated and non-implicated. We drop all funds with missing

ticker symbols. We also drop funds that are in the incubation period --- funds with fewer

than 12 months of observations --- and funds whose TNA is smaller than 5 million USD.

We observe outliers in flows, such as negative flows that are larger than TNA and

positive flows that are five times larger than TNA. We windsorize the sample at the first

and ninety-ninth percentiles for flows to reduce the effect of outliers. The observation

unit is a fund-month. The final sample covers 8,703 funds, of which 1,102 are implicated

funds and 1,003 are non-implicated funds in implicated families. There are 763,072 fund

months from February 1996 to December 2005.

Panels A and B of Table 2 present snapshots of funds in non-implicated

families, and implicated and non-implicated funds in implicated families as of December

2002 and December 2004, respectively. The panels show the number of funds, the mean,

and the total TNA of funds in each group. Sample funds manage 4.7 trillion USD as of

December 2002 and 5.5 trillion USD as of December 2004. Twenty percent and 23% of

total funds under management are controlled by implicated funds in December 2002 and

  17

2004, respectively. This observation does not conflict with the shrinking of implicated

group, because the ratio is **% in June 2003. Furthermore, the average implicated fund is

larger than the average non-implicated fund.

The WRDS database provides information on fund characteristics such as expense

structure (12b-1 fees, rear and front loads), investment style, age (age), investor type

(retail and institutional funds). We use MorningStar’s data on funds’ intial buy restriction

to identify funds for large institutions. We also hand-collect data on some fund

characteristics. We identify whether the parent company is a conglomerate or an asset

management company using SEC EDGAR filings and firm websites. We check whether

funds have a prior history of SEC charges using SEC litigation filings. We compile

montly data for market returns (rm), risk-free rate (rf), the value (SMB), size (HML),

momentum (MOM), and liquidity (LIQ) factors using the Fama French, Momentum, and

Liquidity database from WRDS.

4. Empirical Results

This section discusses the empirical results of our investigation into the four

questions. First, we provide evidence that investors run implicated funds both prior to and

post litigation announcements. The size of silent runs proves both statistically and

economically significant. Second, we investigate whether investors who run prior to

litigation announcements earn higher risk-adjusted returns than do investors who run post

announcements. Third, we analyze how fund and investor characteristics and liquidity of

underlying assets may affect the timing and size of runs and how fund and investor

characteristics, liquidity of underlying assets, and the size of outflows may affect the

costs to investors who run after the public announcement of litigation, or in a relative

  18

sense, the benefits that investors can reap from running early versus late. Finally, using

fund holding and stock return data, we empirically confirm that the cost of running late

is indeed from the fire-sale costs.

4.1 Detecting pre-event runs

We develop two benchmarks to detect silent runs. First, we benchmark flows of

implicated funds against flows of funds in non-implicated families using univariate

analysis. Second, we use multivariate analysis to benchmark flows of implicated funds

against flows estimated using a model of “normal” flows.

Figure 1 plots average monthly flows of funds in non-implicated families, non-

implicated funds in implicated families, and implicated funds from September 2001 to

September 2005. The straight line in Figure 1 indicates September 2003, which is the

month of the first litigation filing. The figure shows that flows of implicated funds are

either higher than or not different from flows of funds in non-implicated families before

April 2003, but consistently lower afterwards. The change of trend starts four months

before the first litigation filing. This pattern suggests that investors ran funds both before

and after the announcement of the first litigation.

Table 3 tests whether the flow differences observed in Figure 1 are statistically

significant. In months prior to September 2003, flows to implicated funds prove

statistically larger than flows to funds of non-implicated families, especially during

September 01 to august 2002. However, the trend reverses in the four months prior to

September 2003. Flows of funds to implicated funds prove smaller (but insignificantly

so) than flows to funds of non-implicated families. In the two years following September

2003, the trend reversal becomes even more pronounced. Flows to implicated funds

  19

prove significantly lower than flows to funds of non-implicated families in all months but

two. Implicated funds that enjoyed large flows up to one year before the onset of

litigations, starts experience runs before September 2003 and statistically significant runs

after September 2003.

Table 3 also compares flows of non-implicated funds in implicated families to

flows of funds in non-implicated families. Flows of the former are significantly lower

than those of the latter in all months in the two years following September 2003.

Investors may regard involvement in these suits as an indication of fund family

managers’ failures to serve investor interests. Consequently, investors punish all funds in

implicated families regardless of whether the fund in question allowed abusive practices

or not, indicating a spill-over effect.

Table 4 estimates the model of flows described in Equation (2) that investigate

whether implicated funds realize abnormal flows around litigation dates. Monthly flows

are regressed on four sets of controls: fund characteristics, past returns, fee structures,

aggregate flows, and on dummy variables for event-window months extending 12 months

before and after litigation announcements7. Table 4 includes six specifications. The first

specification controls for fund characteristics and historic returns. The second and third

specifications introduce controls for fee structure and flow characteristics, respectively.

The last three specifications introduce post-announcement indicators into the first three

specifications. The observation unit is monthly flows from February 1996 to December

2005. Regressions use cluster-robust variance/covariance estimators, where the clusters

are funds.

7 We estimate Equation (2) using fund fixed-effects. The results are available upon request and remain qualitatively the same.

  20

Table 4 confirms the presence of runs (silent and otherwise) that we first detect in

Figure 1 and Table 3. The table shows a significant outflow in implicated funds starting

as early as six months prior to litigation announcements in the fourth specification and as

late as two months in the third and sixth specifications. The significant outflows continue

in the 12 months post litigation announcements. The size of the runs ranges from -23 to -

63 basis points in the month prior to litigation announcements and ranges from -73 to -

103 basis points post announcements. The significant outflows indicate that investors run

implicated funds as soon as they suspect oncoming litigations in the case of pre-litigation

outflows and as soon as litigations are filed in the case of post-litigation outflows.

The four sets of controls prove significant. First, younger and larger firms enjoy

significantly higher flows than do their older and smaller counterparts. Second, investors

chase past returns. Third, funds with high transaction costs (as measured in loads) and

fees (as measured in 12b-1 fees) realize lower flows. Fourth, industry-level and style-

level flows matter. When the industry or the style is enjoying larger flows, so do the

individual funds.

4.2 Benefits of running early versus late

We investigate what benefits exist for investors who run implicated funds prior to

litigation announcements. Pooling all available data on implicated and non-implicated

funds, we estimate a model of “normal” returns to identify return differences of

implicated funds in the months surrounding litigation announcements8. Panel A of Table

8 estimates the models for returns described in Equations (3) through (7). Monthly

8 We employ an alternative approach to detect return differences. We estimate Equations (3) through (7) for each implicated fund seperately. We then test whether the coefficients of pre- and post-event months differ from each other. The differences in coefficients prove significant in the market model and insignificant in the other models.

  21

returns from January 2000 to December 2005 are regressed on dummy variables for

event-window months and the risk factors9. We test for differences in the coefficients of

event-window dummies to detect return differences. All regressions use cluster-robust

variance-covariance estimators where the clusters are the mutual funds.

Panel A of Table 5 shows that investors who run implicated funds post litigation

announcements put up with low returns. The estimates from the market model indicate

that the cost of exiting implicated funds in the six months following litigation

announcements range from 3 basis points in the fifth month to 42 basis points in the

second month. In contrast, investors benefit from exiting implicated funds in three out of

the six months preceding litigation announcements. The results of the other four return

models prove qualitatively similar.

Investors who exit implicated funds before other investors do avoid the lower

returns that investors who exit after litigation announcements suffer from. This result is

consistent with Coval and Stafford’s (2007) argument that prices of underlying assets

become depressed when there is a large volume of asset sales. Table 4 shows that mutual

funds face large outflows following litigation announcements. Mutual funds may suffer

fire-sale costs when they try to liquidate their portfolios to satisfy the high redemption

volume. These fire-sale costs would explain the lower returns observed following

litigation announcements.

Panel B of Table 5 tests the hypothesis that investors benefit from exiting

implicated funds prior to litigation announcements. In the first rows for the five event

windows ranging from one month to five months, the panel shows the difference between

9 We estimate Equations (3) through (7) using fund fixed-effects. The results are available upon request and remain qualitatively the same.

  22

the accumulated coefficients of event-month dummies before and after litigation

announcements. In the second rows, the panel reports the F-statistics for the test that the

difference is equal to 0. For the one-month window, the difference in coefficients pre-

and post-announcements ranges from 58 basis points to 63 basis points. For the six-

month window, the difference in accumulated coefficients pre- and post-announcements

ranges from 107 basis points to 237 basis points. The differences prove economically and

statistically significant.

4.3 Cross sectional difference in runs and returns

The degree of investors’ belief of abusive behavior in funds, size of concerted

redemption once the adverse event is confirmed, and the cost of fire-sale affects their

incentive to run the funds before other investors doing so. We conduct fund-by-fund

estimation on the size of runs and returns difference between runs before and after

litigation announcements to examine these effects. Two indictors are introduced in flow

and returns models (equations 2 to 7), one for six months pre- and the other, post- event.

The coefficients on these two indicators estimate silent runs and runs in the flow analysis

and risk-adjusted returns for the two period in the return analysis. These fund level

estimates are then summarized by groups with classification of whether there is a SEC

charge history, whether the management belongs to financial conglomerates, whether

there is actually 12b1 fees charged, retails vs. institutional funds, retail vs. funds large

institutions, liquidity of underlying assets. For summary of return difference, funds are

also classified by whether the outflows in post event window is above or below median.

  23

Table 6 presents these fund-level estimations with panel A summarizing the

coefficients estimates on the indicators of six months pre- and post- event from the flow

model– estimates of silent runs and runs. The first column presents cross sectional mean

and t-statistics of the silent runs and runs for the full sample, The rest columns in panels

A presents the difference of runs and t-statistics of the difference cross groups. We can

see that abnormal flows in both windows are significantly negative implying both silent

runs and runs in funds involved in litigations. Abnormal flows in six-months before the

litigation are significantly more negative in funds with the SEC charge history or funds

that do not belong to financial conglomerates. Abnormal flows in six-months after the

litigation are significantly more negative in retail funds than funds for large institutions

and funds with liquid assets.

Panel B of table 9 compares the coefficient estimates on the indicators of six

months pre- and post- event from the returns models – estimates of return benefit of silent

runs. The first column of panel B presents the return benefits for the full sample. The rest

columns present the difference of return benefits and t-statistics of the difference cross

groups. We can see that the risk-adjusted returns (alpha) are significantly higher in the

pre-event window than those in the post-event window. The difference in difference of

alphas has little significance across fund characteristics, but significant cross the liquidity

of funds’ underlying assets and amount of flows occur during the post- event window.

These results show both silent runs and runs around the adverse event. Risk-

adjusted returns for investors who withdraw before the information become public are

significantly higher than those for investors who withdraw afterwards. Both runs and

return difference are affected by the fund reputation and the liquidity of underlying

  24

assets, which are consistent with the payoff complementaries and liquidation cost

arguments.

The tests are also conducted with panel approach. We estimate the augmented

version of the flow model described in Equation (2) to determine whether funds differ in

their susceptibility to runs in term of timing and size due to investor clientele, fee

structure, or reputation effect such as whether funds are managed by financial

conglomerates, have a SEC charge history. The augmented models interact the dummy

variables for fund characteristic with every term in the original equation. Table 7

investigates whether fund reputation affects the decision of investors to run. The first two

columns report the results concerning management ownership type and the last two

columns report the results concerning the SEC charge records of the management. Both

sets of regressions use cluster-robust variance-covariance estimators where the clusters

are funds. Each regression generates two sets of coefficients with one for stand-alone

variables and the other for interaction terms.

The runs on implicated funds whose parents are conglomerates prove less

significant both prior to and post filing of litigations. Intuitively, fundsoperating under a

conglomerate may be more reputable since they acquire the backing of the conglomerate

and loss of reputation would hurt the abused fund as well as other businesses of the

conglomerate, therefore,investors may be less suspicious of abusive behavior by funds in

conglomerate families. Further, the reduced size of runs post litigation-announcements

indicates that investors do not punish funds in conglomerate families as severely as funds

in non-conglomerate families even when they learn of abusive behavior. Conglomeration

  25

seems to protect funds from suspicions of abusive practices and from punishments for

abusive practices.

An SEC charge history affects the timing but not the size of the runs. In the three

months prior to litigation announcements, the incremental flows to implicated funds with

SEC charge histories cumulate to -1.63%, whereas in the three months post

announcements, the incremental flows cumulate to 1.70%. A lack of SEC investigations

protects funds from suspicions of abusive practices. However, results indicate that

reputation is lost as soon as investors learn of abusive practices. Even though fund

management may be innocent until proven guilty in the eye of the law, investors seem to

presume guilt as soon as they learn of litigations.

Table 8 analyzes whether investor clientele affects the decision to run and the

setting of the regressions are the same as in table 7. The first three columns report the

results for different investor types because there are two dummies indicating retail funds

and institutional funds indicators, respectively. The last two columns report the results

concerning distribution channels with a dummy indicator for 12-b1 fees. Both sets of

regressions use cluster-robust variance-covariance estimators where the clusters are the

mutual funds.

The information collection and processing skills of investors seem to matter for

detecting abusive behavior prior to such alleged behavior becoming public knowledge.

Investors who are aided by financial advisors (funds with 12b-1 fees) run funds more

both prior to and post filing of litigations. It is also possibly due to self-selection that

active investors who follow the financial news may be more likely to employ financial

  26

advisors. As such, the larger runs might be explained by investor interest rather than the

benefits to retaining financial advisors.

The run size for institutional investors proves significantly lower. This may be

explained by agency problems (James and Karceski 2006). In the late analysis with fund

level estimates, we indentify institutional investors by requiring the fund’s initial

minimum purchase to be $100,000 or above. We find that large institutional do take

action when the funds they invest in allegedly allow abusive practices. This result

supports the agency-problem argument.

Finally, we look at the effect of underlying assets liquidity on the return

difference between withdrawals made prior and post announcements. The benefit of

silent runs is to avoid the liquidation cost (price depression) that investors bear when

funds are forced to sell assets upon revelation of an adverse event. Therefore, investors in

funds with illiquid assets, such as REITs, international assets, or municipal funds have

stronger incentives to run since benefits to running may be greater.

We estimates the augmented version of return model described in Equations (3) through

(7) to investigate whether liquidity of underlying assets affects the return differences

investors realize when they exit before versus after litigation announcements. The

augmented models interact the liquid fund indicator with every term in equations (3)

through (7). Monthly returns from January 2000 to December 2005 are regressed on

dummy variables for event-window months, the risk factors, and the interaction terms.

We report neither the coefficients on the standalone variables or interactions avoid

excessively long tables, although the main information deliverable from these

coefficients are that investors of liquid funds (compared to investors of illiquid funds)

  27

enjoy higher returns in the four months surrounding litigation announcements. In all

specifications of the returns model, returns to illiquid funds are negative up to three

months prior to litigation announcements and remain negative up to four months post

announcements.

Instead, we test for differences in the accumulated coefficients of event-window

dummies and interaction with the liquid indicator,and present the results in table 9. In all

specifications and event-windows, Panel A finds significant return differences between

investors who exit before and after litigation announcements. Panel B shows that the

return differences between investors who exit pre- and post-announcements is less

pronounced in liquid funds. The results support the hypothesis that liquidation cost is

higher in illiquid funds.

To sum up, silent runs are more prominent in funds with bad reputation, illiquid

assets, and institutional investors. These findings are consistent with the payoff

complementary story, because investors are more likely to doubt these funds’ behavior,

withdrawals uponpublic announcement of misbehavior and price depressions during fire-

sale are likely to be larger in these funds.

4.4 Evidence from holding data

Talk about Table 10.

Talk about Table 11.

5. Conclusion

  28

This is the first paper that documents silent runs in mutual funds. We find that

silent runs start as early as six months prior to announcements of litigations. The size of

silent runs before the event becomes public is half the size that it is after. Fund and

investor characteristics such as reputation affect the timing and size of runs. We also find

that investors who run funds prior to litigation announcements realize higher returns than

those who run afterwards, especially in the less-liquid funds, because concerted runs after

the announcement of adverse events trigger significantly fire-sale costs. These results

suggest that the pro-rata distribution of proceeds from asset sales is not sufficient to

prevent silent runs or ensure fairness among investors, since returns to investors differ

across the timing of withdrawals.

The rationale for exiting early has a critically important implication for the

stability of the fund industry. Once the timing of an action matters for payoff, payoff

complementary strategy will prevail. Investors may run funds in the expectation that

other investors will do so. It can amplify the impact of adverse events or random shocks

to fundamentals on financial markets. Mutual funds therefore may face the financial

fragility thatfear of liquidity-dry-up causes runs of liquidity. However, the devastating

consequences that a bank run would confer are not likely to manifest in the fund industry,

since depressed price during fire-sale can be soon recovered as long as the liquidity shock

does not cover all sectors. In case of fund run caused by mispricing set by the funds, it

will stop when price is reset with fair value.

  29

References

J. Bulow, J. Geanakoplos, and P. Klemperer, 1985, Multimarket oligopoly: strategic

substitutes and strategic complements, Journal of Political Economy, 93, 488-

511.

Carhart, Mark M., 1997, On persistence in mutual fund performance, Journal of Finance,

52, 57-82.

Chari, V. V., and Ravi Jagannathan, 1988, Banking panics, information and rational

expectations equilibrium, Journal of Finance, 43, 749-761.

Coval, Joshua and Erik Stafford, 2007, Assets fire sales (and purchases) in equity market,

Journal of Financial Economics, 86, 479-512.

Chen, Qi, Itay Goldstein, and Wei Jiang, 2008, Payoff Complementarities and Financial

Fragility: Evidence from Mutual Fund Outflows, Working Paper, Wharton

School.

Chen, Joseph, Samuel Hanson, Harrison Hong, and Jeremy C. Stein, 2008, Do Hedge

Funds Profit From Mutual-Fund Distress? NBER Working Paper.

Del Guercio, Diane, and Paula A. Tkac, 2001. Star Power: the effect of Morningstar

ratings on mutual fund flows. Working paper, Federal Reserve Bank of Atlanta.

Del Guercio, Diane, and Paula A. Tkac, 2002. The determinants of the flow of funds of

managed portfolios: mutual funds versus pension funds. Journal of Financial and

Quantitative Analysis, 37.

Diamond, Douglas W. and Philip H. Dybvig, 1983, Bank runs, deposit insurance, and

liquidity, Journal of Political Economy, 91, 1983, 401-419.

  30

Dodd, Peter and Jerold B. Warner, 1983. On corporate governance: a study of proxy

contests. Journal of Financial Economics 11, 401-438.

Fama, Eugene F. and Kenneth R. French, 1992, The cross-section of expected stock

returns, Journal of Finance, 67, 427-465.

Fama, Eugene F. and Kenneth R. French, 1993, Common risk factors in the returns on

stocks and bonds, Journal of Financial Economics, 33, 3-56.

Ferson, Wayne E., and Meijun Qian, 2004, Conditional Performance Evaluation,

Revisited. The Research Foundation of CFA Institute. (September).

Ferson, Wayne E., and Vincent A. Warther, 1996, Evaluating Fund Performance in a

Dynamic Market, 1996, Financial Analysts Journal 52, no. 6, pp.20-28.

Greenbaum, Stuart I. and Anjan V. Thakor, 2007 Contemporary Financial

Intermediation, Elsevier Inc.

Gruber, Martin, 1996. Another puzzle: the growth in actively managed mutual funds.

Journal of Finance 51, 783-810.

Hogue, T. and J. Wellman, 2005, Fallout from the Mutual Fund Trading Scandal, Journal

of Business Ethics, 62, 129-139.

Jagadeesh and Titman (1993). N. Jagadeesh and S. Titman , Returns to buying winners

and selling losers: implications for stock market efficiency. Journal of Finance,

48, 65–91.

James, Christopher and Jason Karceski , 2006, Investor Monitoring and Differences in

Mutual Fund Performance, Journal of Banking and Finance, October 2006, Vol.

30, 2787-2808.

  31

Lintner, John, 1965, The valuation of risk assets and the selection of risky investments in

stock portfolios and capital budgets, Review of Economics and Statistics, 47, 13-

37.

Pastor, Lubos, and Robert F. Stambaugh, 2002, Mutual Fund Performance and Seemingly

Unrelated Assets, Journal of Financial Economics, vol. 63, no.3 (March): 315-

349.

Pastor, Lubos, and Robert F. Stambaugh, 2003, Liquidity Risk and Expected Stock

Returns, Journal of Political Economy, vol. 111, no. 3:642–685.

Scholes, Myron, and Joseph Williams, 1977, Estimating Betas from Nonsynchronous

Data, Journal of Financial Economics, Vol. 5, pp307-327.

Sharpe, William F., 1964, Capital assets prices: A theory of market equilibrium under

conditions of risk, Journal of Finance, 19, 425-442.

Sirri, Erik R. and Peter Tufano, 1998. Costly search and mutual fund flows. Journal of

Finance 53, 1589-1622.

Table 1 - List of fund families involved in the trading scandals

A firm is included only if the funds it manages are implicated. Hedge funds, brokerage firms, and other investment banking services are excluded. Some of the allegations are at the state level and some are informal. AG stands for Attorney General; WI stands for Wisconsin; MA stands for Massachusetts; NY stands for New York.

Fund family Practice under Investigation

Regulator involved Initial news date Parent firms

Alliance Bernstein Market timing Internal Probe 9/30/2003 Alliance Capital

Nations Market timing + Late trading

NY State AG 9/3/2003 Bank of America

One Group Market timing NY State AG 9/3/2003 Bank One

Columbia Trading Practice SEC 1/15/2004 Fleet Boston Financial

Federated Market timing + Late trading

SEC/ NASD /NY State AG

10/22/2003 Federated Investors

Franklin Templeton Market timing California AG 9/3/2003 Franklin Resources

Fred Alger & Co

Late trading NY State AG, NY Supreme Court

10/3/2003 Private

Fremont Market timing SEC 11/24/2003 Private

Heartland Advisors Trading Practice + Pricing violation

SEC 12/11/2003 Private

Invesco AIM Market timing SEC/ NY State AG AG/ Colorado AG

12/2/2003 Amvescap PLC

Janus Market timing NY State AG 9/3/2003 Janus Capital Group

Loomis Sayles & Co Market timing Internal Probe 11/13/2003 CDC Assets Management

MFS Market timing SEC 12/9/2003 Sun Life Financial

PBHG Funds market timing SEC/NY State AG 11/13/2003 Old Mutual PLC

Pimco/PEA Capital Market timing California AG 2/13/2004 Allianz Group

Putnam Investments Market timing SEC/ MA State Regulators 9/19/2003 Marsh & McLennan

Scudder Investments Market timing SEC 1/23/2004 Deutsche Bank

Strong Capital Market timing NY State AG/WI State Regulators 9/3/2003 Private

RS Investment Market Timing SEC/NY State AG 3/3/2003 Private

Excelsior Market Timing +late trading

Maryland AG 11/14/2003 Charles Schwab

ING Investment Market Timing + late trading

NY State AG 3/11/2004 ING Group

Evergreen Market Timing MA AG 8/4/2004 Wachovia

Seligman Trading practices +Market Timing

NY State AG 1/7/2004 Private

American Funds Market timing California AG 12/29/2003 Capital Group

Prudential Securities Market timing + late trading

SEC/NASD/NY State AG /MA State Regulators

11/4/2003 Prudential Securities

Sources: Money Management Executive Compilation, January 31, 2004, Wall Street Journal, “Fund Scandal Scorecard” April , 27th 2004, SEC press releases from September 2003 to December 2004. Financial Times 2003-2005. Stanford Law School Library Securities Class Action Clearing House.

  33

Table 2 - Summary statistics: overview of sample funds Panels A and B present snapshots as of December 2002 and December 2004, respectively, for funds in non-implicated families and for implicated and non-implicated funds in implicated families. The table presents total number of funds, average TNA of each fund and total TNA of funds in the three groups. Funds in Non-

implicated Families

Non- implicated Funds in implicated families

Implicated Funds

Panel A: Snap shot at December 2002

Total # of funds 4,664 1,003 1,095

Average TNA of each fund ( million USD) 636 790 867

Total TNA ( million USD) 2,969,569 792,494 949,524

Panel B: Snap shot at December 2004

Total # of funds 4,333 817 1077

Average TNA of each fund ( million USD) 803 871 1,187

Total TNA ( million USD) 3,480,074 712,093 1,277,864

  34

Figure 1 – Plot of flows to funds in non-implicated families, non-implicated funds in implicated families, and implicated funds The figure plots the trends of flows from September 2001 to September 2005 in three fund groups: funds in the non-implicated families, non-implicated funds in the implicated families, and implicated funds. Flowi,t is calculated as [TNAi,t –TNAi,t-1 *(1+Ri,t)] / TNAi,t-1.

Basak: Can you change the blue line into dashed-line? Looking at the print-outs of current one, it is very hard t tell the green from blue (red is fine because it is dark). I think this why the referee says that he cannot differentiate. Let’s change now, since journal graphs are in black and white anyway.

  35

Table 3 - Flows to funds in non-implicated families, implicated and non-implicated funds in implicated families

This table averages monthly flows within non-implicated and implicated families from September 2001 to September 2005. Funds in implicated families are categorized as implicated funds for funds that are named in litigations and as non-implicated for funds that are not named. Event month is the first month litigation was announced: September 2003. Funds within the non-implicated families are benchmarks to test for flow differences against implicated and non-implicated funds in implicated families. ** and * denote significance levels at 1% and 5%, respectively.

Months

from September

2003

Date Non-implicated Families

Implicated Families

Non-implicated Funds

t-stat of difference

Implicated Funds

t-stat of difference

-12 Sep-02 -0.02% -0.55% -3.06** 0.33% 2.17*

-11 Oct-02 0.11% -0.19% -1.68 0.33% 1.31

-10 Nov-02 0.46% 0.68% 1.25 0.18% -1.68

-9 Dec-02 -0.20% -0.88% -3.68** -0.27% -0.41

-8 Jan-03 0.17% -0.25% -2.17* 0.20% 0.19

-7 Feb-03 0.15% -0.41% -3.42** 0.18% 0.15

-6 Mar-03 -0.05% -0.27% -1.31 0.14% 1.14

-5 Apr-03 0.04% -0.70% -4.22** 0.49% 2.78**

-4 May-03 0.16% 0.04% -0.73 0.09% -0.47

-3 Jun-03 0.36% 0.33% -0.13 0.31% -0.33

-2 Jul-03 0.20% 0.19% -0.06 0.00% -1.29

-1 Aug-03 -0.06% -0.49% -2.42** -0.14% -0.50

0 Sep-03 -0.35% -0.92% -3.38** -0.33% 0.15

1 Oct-03 0.12% -0.28% -2.24* -0.33% -2.87**

2 Nov-03 0.21% -0.25% -2.71** -1.09% -8.44**

3 Dec-03 -0.14% -0.97% -4.37** -1.03% -5.44**

4 Jan-04 0.49% -0.41% -4.43** -0.40% -5.04**

5 Feb-04 0.17% -0.41% -3.35** -0.47% -4.30**

6 Mar-04 -0.03% -0.59% -2.97** -0.74% -4.48**

7 Apr-04 -0.42% -1.22% -4.51** -0.93% -3.44**

8 May-04 -0.52% -0.61% -0.49 -1.32% -5.37**

9 Jun-04 -0.35% -0.82% -2.70** -1.11% -5.35**

10 Jul-04 -0.15% -0.74% -3.44** -0.97% -5.85**

11 Aug-04 -0.20% -0.66% -2.85** -0.85% -4.85**

12 Sep-04 -0.35% -1.01% -3.96** -0.72% -2.69**

  36

Table 4 – Detecting silent runs: multivariate analysis of monthly flows The table runs six specifications of the flow model: Flow = a + ∑bj * fund characteristicsj + ∑cj * past returnsj + ∑dj * aggregate flowsj + ∑γj * Event-window dummiesj + ε. The dependent variable is computed as Flowi,t = [TNAi,t –TNAi,t-1 *(1+Ri,t)] / TNAi,t-1. Fund characteristics include size (log of TNA in million USD), age (log of days since first offer date), 12b-1 fees (annual fees paid to financial advisors, measured as percentage of TNA), rear and front loads (charges for purchasing and redeeming shares, as a percentage of TNA). Past returns include cumulative returns in the past one, three, and six months. Aggregate flows include industry- and style-level flows. Industry-level flows are the sum of flows in dollars (TNAi,t –TNAi,t-1

*(1+Ri,t )) to all funds in the sample divided by the sum of lagged TNA (TNAi,t-1). Style-level flows are the sum of flows in dollars to all the funds with the same investment style divided by the sum of lagged TNA. Event-window dummy (n month) equals 1 if it is the nth month to the date of litigation filing and 0 otherwise (n = -1, -2, - - -12, 0, 1, 2, - - -12). Observations are monthly and cover from February 1996 to December 2005. Robust t statistics are in brackets. * indicates significance at 5% and ** significance at 1%. (1) (2) (3) (4) (5) (6)

Dummies for Months -12 to -7 are controlled for.

Dummy (-6 month) -0.21 -0.09 0.09 -0.23* -0.13 0.04 [1.84] [0.75] [0.72] [2.06] [1.08] [0.35] Dummy (-5 month) -0.32** -0.32** -0.01 -0.34** -0.36** -0.06 [2.87] [2.97] [0.12] [3.10] [3.33] [0.54] Dummy (-4 month) -0.16 -0.07 0.07 -0.18 -0.12 0.03 [1.44] [0.62] [0.55] [1.68] [0.96] [0.23] Dummy (-3 month) -0.40** -0.23* 0.02 -0.42** -0.28* -0.02 [3.68] [2.00] [0.17] [3.91] [2.34] [0.20] Dummy (-2 month) -0.44** -0.43** -0.17 -0.47** -0.48** -0.22* [3.82] [4.27] [1.76] [4.04] [4.61] [2.16] Dummy (-1 month) -0.60** -0.52** -0.23* -0.63** -0.57** -0.28* [5.44] [4.61] [2.08] [5.66] [4.93] [2.45] Event month 0 -1.03** -0.95** -0.73** [9.80] [8.56] [6.62]

Dummy (+1 month)   -1.77** -1.97** -1.76**   [12.71] [11.80] [10.54] Dummy (+2 month)   -1.58** -1.38** -1.07**   [15.71] [13.50] [10.38] Dummy (+3 months)   -1.32** -1.33** -1.14**   [12.39] [12.13] [10.30] Dummy (+4 months)   -1.05** -0.99** -0.74**   [11.89] [11.68] [8.75] Dummy (+5 month)   -1.04** -0.93** -0.75**   [12.85] [10.53] [8.63] Dummy (+6 month)   -0.77** -0.73** -0.48**   [9.08] [9.39] [6.27] Dummy (+7 month)   -0.99** -0.73** -0.48**

  37

  [12.35] [9.26] [6.11] Dummy (+8 month)   -0.89** -0.73** -0.53**   [11.39] [9.99] [7.49] Dummy (+9 month)   -0.78** -0.67** -0.53**   [9.91] [8.86] [7.05] Dummy (+10 month)   -0.72** -0.77** -0.57**   [8.22] [10.47] [7.71] Dummy(+11 month)   -0.51** -0.55** -0.40*   [5.14] [3.40] [2.46] Dummy (+12 month)   -0.61** -0.77** -0.58**   [8.38] [7.72] [5.76] Age (LogAge) -1.04** -1.18** -1.14** -1.03** -1.16** -1.12** [45.28] [36.87] [35.75] [44.82] [36.30] [35.33] Size (logTNA) 0.26** 0.23** 0.22** 0.26** 0.23** 0.23** [33.21] [21.49] [20.91] [33.36] [21.50] [20.94] Return in the 1.78** 1.40** 0.93** 1.77** 1.39** 0.95** last month [10.55] [7.28] [4.81] [10.50] [7.26] [4.93] Cumulative Returns 0.58** 0.27* 0.13 0.55** 0.23 0.1 in the past 3 months [5.27] [2.07] [0.99] [4.96] [1.77] [0.76] Cumulative Returns 4.55** 5.21** 4.41** 4.63** 5.29** 4.48** in the past 6 months [45.32] [41.25] [33.94] [45.65] [41.33] [34.07] Front+ Rear Load -0.03** -0.02** -0.03** -0.02** [4.44] [2.82] [4.27] [2.75] Actual 12b-1 Fess -0.71** -0.60** -0.67** -0.57** [13.13] [10.96] [12.19] [10.33] Industry-Normalized Flow 0.05* 0.03 [2.00] [1.13] Style-Normalized Flow

0.54** 0.53** [31.18] [31.11] Constant 6.93** 8.62** 8.08** 6.86** 8.47** 7.99** [39.85] [35.59] [33.47] [39.45] [34.93] [33.03] Observations 660,317 355,811 355,811 660,317 355,811 355,811 Adjusted R-squared 3.60% 5.96% 7.61% 3.70% 6.09% 7.69%

  38

Table 5: Fund returns before and after litigation announcements Pooled regressions using CAPM, Fama-French, Cahart, and William and Scholes (1976), and Pastor Stambaugh (2000) models are run. Observations are from January 1, 2000 to December 31, 2005. The dependent variable is monthly fund returns (in %). A dummy (n month) equals 1 if it is the nth month before (-n) or after (+n) litigations are filed. For other months and non-indicted funds, the dummy takes on the value 0, n = -1, -2, - - - -6, 0 , 1, 2, - - - 6. Panel A presents the regression results. Robust t statistics are in brackets. * indicates significance at 5% and ** significance at 1%. Panel B tests for differences in the accumulated abnormal returns between D-n and D+n.

Market model

Fama-French

Carhart four

factors

Market model with

lagged returns

Market model with

liquidity factor

Panel A: Regression results

Dummy (-6 month) 0.36** 0.29** 0.29** 0.32** 0.37** [4.41] [3.57] [3.65] [3.91] [4.56] Dummy (-5 month) 0.46** 0.31** 0.32** 0.39** 0.47** [5.03] [3.37] [3.52] [4.21] [5.11] Dummy (-4 month) 0.00 -0.18* -0.18* -0.08 0.00 [0.02] [2.35] [2.32] [0.95] [0.04] Dummy (-3 month) -0.01 -0.12 -0.12 -0.07 0.00 [0.09] [1.53] [1.52] [0.90] [0.03] Dummy (-2 month) -0.08 -0.21** -0.21** -0.13 -0.08 [1.11] [2.80] [2.80] [1.70] [1.14] Dummy (-1 month) 0.54** 0.45** 0.45** 0.48** 0.57** [7.60] [6.24] [6.21] [6.67] [8.04] Dummy (0 month) 0.34** 0.30** 0.29** 0.31** 0.36** [4.89] [4.17] [4.09] [4.46] [5.21] Dummy (+1 month) -0.07 -0.13 -0.14* -0.12* -0.07 [1.26] [2.27]* [2.32] [2.08] [1.15] Dummy (+2 month) -0.40** -0.26** -0.25** -0.46** -0.41** [6.41] [4.25] [4.17] [7.41] [6.47] Dummy (+3 months) -0.15** -0.14** -0.14** -0.19** -0.15** [2.93] [2.66] [2.73] [3.70] [2.92] Dummy (+4 months) -0.08* -0.05 -0.05 -0.12** -0.11* [1.98] [1.23] [1.27] [2.75] [2.53] Dummy (+5 month) 0.00 0.11 0.12 -0.02 -0.03 [0.02] [1.78] [1.85] [0.29] [0.54] Dummy (+6 month) -0.19** -0.02 -0.02 -0.18** -0.23** [3.28] [0.37] [0.27] [3.10] [3.91]

Continue on the next page,

  39

Table 5, continued.

Market Returns 0.49** 0.51** 0.51** 0.49** 0.49** [77.40] [83.19] [85.18] [77.53] [76.48] SMB 0.08** 0.07** [29.07] [30.28] HML 0.07** 0.07** [24.28] [23.71] Momentum 0.00** [3.41] Lagged market returns 0.02** [21.69] Liquidity factor 0.01** [21.41] Constant 0.10** -0.03** -0.03** 0.11** 0.15** [18.94] [8.85] [8.76] [20.02] [22.14] Observations 473,508 473,508 473,508 466,562 473,508 R-squared 30.73% 31.27% 31.28% 31.02% 30.77% Panel B: Performance Difference

Dummy(-1 month) 0.62 0.58 0.58 0.60 0.64 - Dummy (+1 month) 54.69 48.93 49.09 50.24 59.47 Accumulate (-1 to -2) 0.94 0.63 0.63 0.94 0.97 - Accumulate (+1 to +2) 71.57 33.44 33.11 70.28 75.01 Accumulate (-1 to -3) 1.09 0.65 0.65 1.06 1.12 - Accumulate (+1 to +3) 65.72 25.27 25.43 63.53 68.10

  40

Table 6: Cross sectional difference in silent runs, runs, and return benefits This table presents the summary results of individual fund estimates. The analysis consists of two steps. In the first step, we run time series regressions of flows as in equation (2) in panel A and of returns as in equations (3) to (7) in panel B for each fund. In the second step, we compare these fund level estimates across groups. The fund groups are classified according to their SEC charge history, whether the management belongs to a financial conglomerates, whether there is actually 12b1 fees charged, retails vs. institutional funds, retail vs. funds large institutions, liquidity of underlying assets, and whether the outflows in post event window is above or below median.

Panel A presents the cross sectional mean and t-statistics of the coefficients on the indicators of six months pre- and post- event from the flow model– estimates of silent runs and runs. Panel B presents the difference in cross sectional mean and t-statistics of the coefficient on the indicators of six months pre- and post- event from the returns models – estimates of return benefit of silent runs. Panel A: Mean and t-statistics of the silent runs and runs for the full sample, and difference cross groups

Full Sample

Charge History (No–Yes)

Ownership (Other – Conglo-merates)

12-b1 fees (Without-With)

Clientele (Retail – Institution)

Clientele (Retail – large Institution)

Liquidity of Underlying Assets (Illiquid –Liquid)

Flows (silent runs) -0.42 0.56 -0.57 -0.07 0.09 0.48 0.15

(-6 to -1) -18.88 3.26 -3.08 -0.26 0.22 1.26 0.56

Flows (runs) -1.44 0.01 -1.03 0.29 -0.75 -0.69 0.62

(+1 to +6) -5.79 0.06 -5.39 0.95 -1.69 -1.88 2.29

Panel B: Mean and t-statistics of the return benefits of silent runs over runs for the full sample and difference cross groups.

Alpha (+1 to +6) – (-6 to -1)

Full Sample

Charge History (No–Yes)

Ownership (Other – Conglo-merates)

12-b1 fees (Without-With)

Clientele (Retail – Institution)

Clientele (Retail – large Institution)

Liquidity of Underlying Assets (Illiquid –Liquid)

Outflow (Large –Small)

Market model 0.32 0.09 -0.05 -0.05 0.14 0.23 0.25 0.19

13.34 1.66 -0.83 -0.63 0.96 2.5 3.41 4.12

Fama-French 0.09 0.12 -0.03 -0.07 0.05 0.1 0.21 0.1

4.13 2.26 -0.52 -1.05 0.47 1.15 3.39 2.43

Carhart Four 0.12 0.15 0.01 -0.07 0.07 0.16 0.23 0.08

Factors 6.03 3.16 0.26 -1.22 0.65 1.77 4.32 2.06

Market and lagged 0.28 0.1 -0.04 0.02 0.19 0.15 0.26 0.18

market returns 12.15 1.78 -0.72 0.29 1.45 1.62 3.63 3.95

Market and 0.34 0.09 -0.07 -0.08 0.13 0.3 0.27 0.22

liquidity factor 13.54 1.53 -1.19 -1.07 0.87 3.12 3.42 4.39

  41

Table 7: The effect of reputation on silent runs

This table runs the augmented flow model (Flow = a + ∑bj * fund characteristicsj + ∑cj * past returnsj + ∑dj

* aggregate flowsj + ∑γj * Event-window dummiesj + ε), with characteristics dummy to interact with every term in the model. The first characteristics dummy equals one if funds managed by conglomerates and zero if stand-alone asset-management companies. The second characteristics dummy equals one if funds whose managements have been charged by SEC in the past eight years and zero if not. The dependent variable is computed as Flowi,t = [TNAi,t –TNAi,t-1 *(1+Ri,t)] / TNAi,t-1. Controls for fund characteristics include size (log of TNA in million USD), age (log of days since first offer date), 12b-1 fees (annual fees paid to financial advisors, measured as percentage of TNA), rear and front loads (charges for purchasing and redeeming shares, as a percentage of TNA). Past returns include cumulative returns in the past one, three, and six months. Aggregate flows include industry- and style-level flows. Industry-level flows are the sum of flows in dollars (TNAi,t –TNAi,t-1 *(1+Ri,t )) to all funds in the sample divided by the sum of lagged TNA (TNAi,t-1). Style-level flows are the sum of flows in dollars to all the funds with the same investment style divided by the sum of lagged TNA. Event-window dummy (n month) equals 1 if it is the nth month before or after the litigation is filed and 0 otherwise (n = -1, -2, - - -12, 0, 1, 2, - - -12). Observations are monthly and cover from February 1996 to December 2005. Robust t statistics are in brackets. * indicates significance at 5% and ** significance at 1%.

Regression with

conglomerate indicators Regression with charge-

history indicators

Stand-alone variable

Variables interacted with conglomerate indicators

Stand-alone variable

Variables interacted with charge- history indicators

Dummies for Months -12 to -7 are controlled for.

Dummy (-6 month) -0.30* 0.74** 0.10 -0.54* [2.32] [2.62] [0.63] [2.46] Dummy (-5 month) -0.37** 0.90** 0.11 -0.57* [2.89] [3.12] [0.74] [2.35] Dummy (-4 month) -0.38* 0.37 -0.13 -0.30 [2.56] [1.63] [1.02] [1.12] Dummy (-3 month) -0.66** 0.74** -0.24 -0.54* [5.82] [2.74] [1.79] [2.55] Dummy (-2 month) -0.60** 0.57* -0.24 -0.53* [4.29] [2.11] [1.47] [2.40] Dummy (-1 month) -0.64** 0.50 -0.28 -0.56* [4.77] [1.74] [1.74] [2.52] Event month 0 -1.12** 0.43 -0.99** 0.01 [8.98] [1.68] [6.78] [0.07] Dummy (+1 month) -2.30** 1.41** -2.30** 1.24** [12.15] [4.34] [10.71] [4.40] Dummy (+2 month) -1.87** 1.25** -1.61** 0.29 [14.79] [5.21] [10.79] [1.43] Dummy (+3 months) -1.69** 1.48** -1.30** 0.17

  42

[12.98] [5.27] [9.08] [0.66] Dummy (+4 months) -1.35** 1.09** -1.03** 0.03 [12.01] [5.47] [8.64] [0.16] Dummy (+5 month) -1.22** 0.66** -1.02** 0.03 [11.34] [3.32] [9.18] [0.17] Dummy (+6 month) -1.01** 0.93** -0.81** 0.25 [10.50] [5.30] [7.64] [1.54] Dummy (+7 month) -1.01** 0.76** -0.96** 0.46** [10.06] [4.18] [8.43] [2.88] Dummy (+8 month) -0.95** 0.55** -0.98** 0.54** [10.82] [3.06] [9.78] [3.49] Dummy (+9 month) -0.87** 0.39 -0.86** 0.32 [10.61] [1.77] [8.36] [1.89] Dummy (+10 -0.71** 0.18 -0.62** -0.14 month) [7.76] [0.66] [4.96] [0.74] Dummy(+11 -0.50** 0.08 -0.46** 0.03 month) [3.57] [0.36] [4.08] [0.09] Dummy (+12 -0.68** 0.04 -0.66** 0.03 month) [7.14] [0.24] [7.42] [0.15] Age (LogAge) -1.00** -0.13* -1.05** 0.00 [26.64] [2.31] [28.73] [0.02] Size (logTNA) 0.22** 0.06** 0.23** 0.03 [18.03] [3.09] [19.02] [1.83] Return in the 1.96** -2.11** 0.21 1.70** last month [7.13] [5.46] [0.81] [4.32] Cumulative Returns 0.10 0.48 0.11 0.36 in the past 3 months [0.55] [1.82] [0.65] [1.36] Cumulative Returns 4.52** -0.89** 4.32** -0.34 in the past 6 months [26.66] [3.68] [25.35] [1.38] Front+ Rear Load 0.00 -0.05** 0.00 -0.03* [0.26] [3.69] [0.23] [2.27] Industry-Normalized 0.03 0.01 0.06 -0.03 Flow [1.11] [0.34] [1.84] [0.59] Style-Normalized 0.53** 0.03 0.50** 0.09** Flow [25.42] [0.84] [24.10] [2.81] Constant 6.77** 0.59 7.16** -0.38 [23.41] [1.39] [25.16] [0.89] Observations 447,219 447,219 Adjusted R-squared 6.14% 6.11%

  43

Table 8 - The effect of investor type on runs

This table runs the augmented flow model (Flow = a + ∑bj * fund characteristicsj + ∑cj * past returnsj + ∑dj

* aggregate flowsj + ∑γj * Event-window dummiesj + ε) with characteristics indicator interact with all terms in the model. The first set of characteristics indicator include retail funds and institutional funds, The second characteristic indicator equals one if funds are distributed through financial advisory services and zero if not. The dependent variable is computed as Flowi,t = [TNAi,t –TNAi,t-1 *(1+Ri,t)] / TNAi,t-1. Controls for fund characteristics include size (log of TNA in million USD), age (log of days since first offer date), 12b-1 fees (annual fees paid to financial advisors, measured as percentage of TNA), rear and front loads (charges for purchasing and redeeming shares, as a percentage of TNA). Past returns include cumulative returns in the past one, three, and six months. Aggregate flows include industry- and style-level flows. Industry-level flows are the sum of flows in dollars (TNAi,t –TNAi,t-1 *(1+Ri,t )) to all funds in the sample divided by the sum of lagged TNA (TNAi,t-1). Style-level flows are the sum of flows in dollars to all the funds with the same investment style divided by the sum of lagged TNA. Event-window dummy (n month) equals 1 if it is the nth month before or after the litigation is filed and 0 otherwise (n = -1, -2, - - -12, 0, 1, 2, - - -12). Observations are monthly and cover from February 1996 to December 2005. Robust t statistics are in brackets. * indicates significance at 5% and ** significance at 1%.

Regression with retail and institutional fund indicators

Regression with 12b-1 fees indicators

Stand-alone variable

Variable interacted with indicator for retail fund

Variable interacted with indicator for institutional fund

Stand-alone variable

Variable interacted with indicator for 12b-1 fees

Dummies for Months -12 to -7 are controlled for.

Dummy (-6 month) -0.46 0.13 0.60* 0.18 -0.27 [0.92] [0.27] [2.43] [0.79] [1.04] Dummy (-5 month) -0.13 -0.24 0.49 0.32 -0.52 [0.31] [0.64] [1.85] [1.32] [1.96] Dummy (-4 month) 0.10 -0.36 0.37 0.21 -0.33 [0.27] [1.00] [1.51] [0.96] [1.31] Dummy (-3 month) -0.15 -0.20 0.16 -0.23 0.06 [0.34] [0.46] [0.68] [1.04] [0.24] Dummy (-2 month) 0.09 -0.49 0.21 0.28 -0.65* [0.20] [1.09] [0.88] [1.01] [2.20] Dummy (-1 month) -0.05 -0.65 0.48 -0.14 -0.28 [0.10] [1.36] [1.94] [0.59] [1.03] Event month 0 -0.67 -0.49 0.67** -0.50* -0.37 [1.53] [1.14] [2.88] [2.26] [1.50] Dummy (+1 month) -1.84** -0.32 1.20** -0.76** -1.15** [4.62] [0.93] [4.03] [3.04] [3.82] Dummy (+2 month) -1.22** -0.48 0.71** -1.26** 0.05 [3.32] [1.40] [3.12] [5.69] [0.18] Dummy (+3 months) -1.17** -0.48 1.07** -0.62** -0.66** [3.16] [1.38] [4.69] [2.73] [2.60] Dummy (+4 months) -0.70 -0.53 0.74** -0.58** -0.31 [1.85] [1.43] [3.93] [2.80] [1.39]

  44

Dummy (+5 months) -0.54 -0.62 0.51** -0.72** -0.17 [1.63] [1.93] [2.94] [4.25] [0.88] Dummy (+6 months) -0.81* 0.01 0.53** -0.26 -0.35 [2.31] [0.02] [2.82] [1.31] [1.64] Dummy (+7 months) -0.90** 0.22 -0.04 -0.76** 0.16 [2.59] [0.66] [0.21] [4.37] [0.81] Dummy (+8 months) -1.51** 0.77* 0.33* -0.62** -0.04 [4.45] [2.34] [2.01] [3.44] [0.19] Dummy (+9 months) -1.10** 0.30 0.47** -0.43* -0.23 [3.16] [0.88] [2.89] [2.42] [1.20] Dummy (+10 months) -0.55 -0.17 0.29 -0.18 -0.52* [1.21] [0.38] [1.83] [0.93] [2.56] Dummy(+11 months) -0.31 -0.20 0.19 -0.18 -0.36 [0.99] [0.70] [0.85] [1.48] [1.76] Dummy (+12 months) -0.50 -0.16 0.28 -0.31** -0.42** [1.96] [0.68] [1.83] [3.09] [2.93] Age (LogAge) -0.82** 0.03 -0.07 -0.90** -0.19** [9.91] [0.36] [1.13] [32.28] [4.29] Size (logTNA) 0.17** 0.05 0.07** 0.26** -0.01 [6.11] [1.88] [3.39] [26.23] [0.41] Return in the 0.55 1.18 0.59 1.60** -0.64 last month [0.77] [1.69] [0.98] [5.76] [1.87] Cumulative Returns -1.38** 2.58** 1.69** 0.24 0.16 in the past 3 months [2.64] [5.10] [4.12] [1.41] [0.75] Cumulative Returns 1.87** 1.96** -0.52 2.97** 1.61** in the past 6 months [4.67] [5.05] [1.57] [18.20] [7.50] Front+ Rear Load 0.05 -0.06* -0.03* 0.01 -0.07** [1.77] [2.17] [2.45] [1.21] [5.04] Industry-Normalized 0.20** -0.26** -0.27** 0.03 0.01 Flow [2.85] [3.72] [4.27] [1.25] [0.19] Style-Normalized 0.35** 0.05 0.14** 0.61** -0.10** Flow [7.01] [1.12] [3.45] [37.85] [4.05] Constant 5.46** -0.48 0.30 5.74** 1.72** [8.49] [0.76] [0.59] [27.10] [5.09] Observations 452,939 660,317 Adjusted R-squared 4.91% 5.35%

  45

Table 9 - The effects of asset liquidity on returns

This table runs the augmented return model: CAPM, Fama-French, Carhart, William and Scholes (1976) and Pastor and Stambaugh (2000), with liquidity indictor interacting with all terms in the model. Each regression generates two sets of coefficients: one set for the stand-alone variables and the other set for the interactions of variables and the indicator for liquid funds. We categorize growth-income and money-market funds as liquid funds and global equity, bond funds, municipal funds, and others (such as Ginnie Mae) as illiquid funds. The dependent variable is monthly fund excess returns in percentage. Event-window dummy (n month) equals 1 if it is the nth month before or after the litigation is filed and 0 otherwise (n = -1, -2, - - -12, 0, 1, 2, - - -12). Observations are monthly and cover from February 1996 to December 2005. Robust t statistics are in brackets. * indicates significance at 5% and ** significance at 1%. Panel A tests for differences in the accumulated coefficients between D-n and D+n. Panel B tests the difference in the accumulated coefficients between D-n and D+n interacted with the liquid indicator.

Market model

Fama-French

Carhart four factors

Market model with

lagged market returns

Market model with

liquidity factor

Panel A: Performance Difference (Stand-alone only) Dummy(-1 month) 1.10** 1.08** 1.09** 1.08** 1.13** - Dummy (+1 month) [10.27] [10.06] [10.11] [9.88] [10.56]Accumulate (-1 to -2) 1.35** 1.08** 1.06** 1.37** 1.38** - Accumulate (+1 to +2) [8.51] [6.97] [6.94] [8.60] [8.60]Accumulate (-1 to -3) 1.37** 0.98** 0.98** 1.36** 1.41** - Accumulate (+1 to +3) [6.74] [4.94] [5.06] [6.67] [6.79]Accumulate (-1 to -4) 1.56** 0.97** 1.00** 1.52** 1.62** - Accumulate (+1 to +4) [6.23] [3.95] [4.15] [6.06] [6.31]Accumulate (-1 to -5) 2.07** 1.23** 1.32** 1.99** 2.17** - Accumulate (+1 to +5) [7.17] [4.35] [4.77] [6.85] [7.36]Accumulate (-1 to -6) 2.53** 1.49** 1.59** 2.41** 2.68** -Accumulate (+1 to +6) [8.01] [4.83] [5.21] [7.56] [8.39]Panel B: Performance Difference (Interacted with liquid indicator)

Dummy(-1 month) 0.39** 0.35** 0.35** 0.39** 0.40** - Dummy (+1 month) [2.79] [2.65] [2.60] [2.77] [2.82]Accumulate (-1 to -2) 0.01 -0.02 -0.02 0.00 0.02 - Accumulate (+1 to +2) [0.00] [0.00] [0.10] [0.00] [0.10]Accumulate (-1 to -3) 0.21 0.24 0.23 0.18 0.22 - Accumulate (+1 to +3) [0.48] [0.61] [0.58] [0.42] [0.50]Accumulate (-1 to -4) 0.43 0.46 0.44 0.38 0.45 - Accumulate (+1 to +4) [0.92] [1.08] [1.01] [0.82] [0.94]Accumulate (-1 to -5) 2.07** 2.00** 1.95** 1.97** 2.11** - Accumulate (+1 to +5) [4.95] [5.18] [4.99] [4.74] [4.95]Accumulate (-1 to -6) 3.56** 3.39** 3.34** 3.43** 3.63** -Accumulate (+1 to +6) [8.93] [8.86] [8.67] [8.62] [8.99]

  46

Table 10: Implicated fund holdings and firm CARs       Held by at least 

one implicated fund 

No holding by implied funds  t‐statistic  p‐value      

Firms for which aggregated change in holdings is negative 

CAR (Sep 2‐5)  ‐1.82% ‐0.11% ‐5.73  0.00

CAR (Sep 8‐12)  ‐1.11% ‐0.21% ‐3.57  0.00

CAR (Sep 15‐19)  ‐1.24% ‐0.96% ‐1.17  0.24

CAR (Sep 22‐26)  0.78% 0.16% 2.32  0.02

CAR (Sep 2‐26)  ‐3.39% ‐1.11% ‐4.28  0.00

Number of observations  765 769      

Firms for which aggregated change in holdings is positive 

CAR (Sep 2‐5)  ‐1.97% 0.64% ‐9.39  0.00

CAR (Sep 8‐12)  ‐0.84% ‐0.23% ‐2.58  0.01

CAR (Sep 15‐19)  ‐0.98% 0.07% ‐3.97  0.00

CAR (Sep 22‐26)  0.63% ‐0.54% 4.26  0.00

CAR (Sep 2‐26)  ‐3.15% ‐0.04% ‐5.81  0.00

Number of observations  814 832      

  47

Table 11: Fund flows and firm CARs Panel A:    Means  Number of observations 

T‐test for the difference between holdings of implied funds with negative flows and others 

Holdings of all funds 

Holdings of implicated funds with negative flows 

Holdings of all other funds, (including implicated funds that have positive flows) 

Holdings of all funds 

Holdings of implied funds with negative flows 

Holdings of all other funds 

   T‐statistic  P‐value

CAR (Sep 2‐5)  ‐0.78%  ‐2.28% ‐0.48% 3,230 530  2700  6.50 0.00

CAR (Sep 8‐12)  ‐0.59%  ‐1.04% ‐0.50% 3,229 530  2699  2.30 0.02

CAR (Sep 15‐19)  ‐0.75%  ‐1.28% ‐0.64% 3,225 529  2696  2.58 0.01

CAR (Sep 22‐26)  0.24%  1.39% 0.01% 3,223 529  2694  ‐5.36 0.00

CAR (Sep 2‐26)  ‐1.87%  ‐3.21% ‐1.61% 3,230 530  2700  3.10 0.00

Panel B: Table 10 - Summary statistics: CARs of firms held in mutual fund portfolios

All firms 

Number of observations 

Firms for which change in holdings (aggregated across all funds) is: 

T‐test for the difference between positive and negative change in holdings 

   Negative  Positive  t‐statistic  p‐value 

CAR (Sep 2‐5)  ‐0.78%  3,230  ‐0.96%  ‐0.61%  ‐1.69  0.09 

CAR (Sep 8‐12)  ‐0.59%  3,229  ‐0.66%  ‐0.52%  ‐0.80  0.42 

CAR (Sep 15‐19)  ‐0.75%  3,225  ‐1.10%  ‐0.43%  ‐3.70  0.00 

CAR (Sep 22‐26)  0.24%  3,223  0.47%  0.02%  2.33  0.02 

CAR (Sep 2‐26)  ‐1.87%  3,230  ‐2.25%  ‐1.54%  ‐1.86  0.06