brokers

Document Sample
brokers Powered By Docstoc
					               The Economics of Mutual-Fund Brokerage:
         Evidence from the Cross Section of Investment Channels


                                          By

                                Susan Christoffersen
                            McGill University and CIRANO

                                    Richard Evans
                                    Boston College

                                   David Musto*
                              University of Pennsylvania



                              This Draft: January 12, 2006




__________________________________
*
 Christoffersen gratefully acknowledges financial support from IFM2 and the Canadian
Securities Institute Research Foundation. The authors also wish to thank the Q-group for
their financial support of this project. We are grateful for comments from Dan
Bergstresser, John Chalmers, Peter Christoffersen, Marcin Kacperczyk, Tim Simin, and
Lu Zheng and seminar participants at Penn State University, Boston College, the Mitsui
Life Conference at the University of Michigan, 2005 Q-Group Meetings in Carlsbad,
2005 NFA Meetings and the 2006 AFA Meetings. Extensive research assistance was
needed in matching databases from Eric Turner, Jerome Grenier, John Bromley, and
Thibault Webanck. Any remaining errors are our responsibility. Contact: David K.
Musto, musto@wharton.upenn.edu, Phone (215) 898-4239, FAX 215-898-6200, Finance
Department, 3620 Locust Walk, Philadelphia PA, 19104-6367..
               The Economics of Mutual-Fund Brokerage:
         Evidence from the Cross Section of Investment Channels




                                      ABSTRACT

Retail investors often lack investment expertise. Mutual-fund brokers can help, but their
incentives are mixed so it is an empirical question what value they add, both for
consumers and for fund families. Investors pay more to invest through unaffiliated
brokers than captive brokers, and while unaffiliated brokers add more value to
redemptions, captive brokers add more value to inflows. No-load investors are less likely
to sell their poor-performing funds and more likely to sell their winning funds, consistent
with a disposition effect. Fund families benefit from a captive salesforce through
recapture of redemptions, but also suffer through cannibalization of inflows.




                                            1
1.       Introduction

         U.S. consumers may be poorly prepared for the investment choices they must

make. Few have any training in the area, but the rise in household wealth gives them

more to invest, while the rise in life expectancy and the decline in traditional pensions

raise the stakes for their choices. It seems odd, in this context, that consumers do not

choose to buy average performance for almost nothing, via index funds, but even this

simple approach requires expensive education to appreciate, and few provide such

education below cost. Education is available from mutual-fund brokers, but they expect

adequate compensation, and like all brokers, they can have responsibilities and incentives

on both sides of the transactions they facilitate. In this paper we look at the role of

mutual-fund brokers from both sides: from the investor side, as sources of good advice,

and from the fund side, as sources of profitable new accounts. A new database lets us

address the question, does the value of brokers to investors and funds depend on their

incentives to deliver for them?

         The new database is the history of electronic filings by U.S. funds. In these

filings we see not only monthly inflows and redemptions, but also the loads arising from

the inflows, broken down by who gets them: captive brokers, unaffiliated brokers, or the

underwriter – i.e., the fund family itself. The distinction between captive and unaffiliated

brokers is the key1, as it separates two potentially disparate qualities of advice, the former



1
 The difference between captive or affiliated and unaffiliated brokers is central to the empirical analysis of
the paper. The actual definition of ‘affiliated’ from the Investment Company Act can be found in Section
2.3: "Affiliated person" of another person means (A) any person directly or indirectly owning, controlling,
or holding with power to vote, 5 per centum or more of the outstanding voting securities of such other
person; (B) any person 5 per centum or more of whose outstanding voting securities are directly or
indirectly owned, controlled, or held with power to vote, by such other person; (C) any person directly or
indirectly controlling, controlled by, or under common control with, such other person; (D) any officer,
director, partner, copartner, or employee of such other person; (E) if such other person is an investment


                                                      2
potentially skewed toward the interests of the fund family, away from the investor. To

these two investment channels we add a third by including no-load funds, representing

investment decisions made without either kind of broker. With this novel data the value

that brokers add is examined by analyzing how and why inflows and redemptions

through these channels differ.

          For an investor, the important difference is likely to be performance. Is

performance good after money flows in? And similarly, is performance bad after money

leaves? The empirical literature documents performance persistence both near the top

and near the bottom of a period’s performers (Hendricks, Patel and Zeckhauser, 1993,

Brown and Goetzmann, 1995, Carhart, 1997), so one way to address these questions are

to ask both whether top returns predict inflows, and whether bottom returns predict

redemptions. To explain why some investors do not redeem from bad performers, the

literature posits a population of disadvantaged, “sleeping” investors (Gruber, 1996,

Christoffersen and Musto, 2002, Berk and Xu, 2004); intuitively this would be a larger

problem for investors with no brokers, and therefore no agents to wake them up, and also

those with captive brokers, who may find it more difficult to reinvest them within the

family.

          Another way to address these questions is simply to relate flows to subsequent

performance. We know from earlier such tests that net flows predict performance

(Gruber, 1996, Zheng, 1999), and that brokered flows underperform direct flows

(Bergstresser, Chalmers and Tufano, 2004). The question for the new data is whether the

difference between brokers’ incentives passes through to the flows they advise.

company, any investment adviser thereof or any member of an advisory board thereof; and (F) if such other
person is an unincorporated investment company not having a board of directors, the depositor thereof.



                                                   3
        At the other end of fund flows, the funds themselves, the key question is what

benefits fund families get from using different channels. Families benefit when they

recapture redemptions from one fund with inflows to others, and they suffer when

inflows to one fund cannibalize inflows to others. Compared to unaffiliated brokers,

captive brokers likely find it easier to attract money from other funds in the family, rather

than from other families, and more rewarding to direct money to other funds in the

family, rather than to other families, so the interest is in whether or not this is evident in

flows. While we cannot test directly on intrafamily flows, we can test indirectly by

comparing the flows of a fund with the reverse flows of the rest of the family. To address

recapture, we test whether the redemptions caused by a fund’s performance correlate

positively with excess inflows to the rest of its family, and to address cannibalization we

test whether excess inflows to a fund correlate negatively with excess inflows to the rest

of its family.

        These results build on the recent work by Bergstresser, Chalmers and Tufano

(2004), who document a generally negative experience investing through brokers. We

extend this research by distinguishing brokers by their incentives, distinguishing flows by

their direction, and also by calculating loads directly from dollars paid.

        The paper is in five sections. Section 2 describes the data, Section 3 provides an

overview of flows, Section 4 contains the main empirics and Section 5 summarizes and

concludes.

2.      Data

        To assess the value of a channel both to the investor and to the fund family, we

need data that separates investors’ inflows from their outflows and that identifies the




                                               4
channel through which the money arrives. Account-level databases such as those

employed by Barber, Odean and Zheng (2004) and Johnson (2004) are possibilities but

they contain limited, non-random samples of investors and funds. These data also end in

the late 1990s when returns were at their peak, so they do not reveal investors’ behavior

in a market downturn. Instead, we use publicly available data from the SEC N-SAR

electronic filings from the period 1996 to 2003. We match these filings to the CRSP

database which gives us a majority of funds in this time period, including both the bull

and bear markets. Reuter (2004) also matches CRSP with N-SAR files but does not

analyze the disaggregated flow data we consider here. O’Neal (2004) analyzes purchases

and redemptions of annual flows using data from the Form 485-B but he does not have

information on the payments to broker channels and focuses on the 200 largest equity

funds.

         For this paper, we use the N-SAR semi-annual reports. In 1994, the SEC started

to make this filing a requirement for a subsample of the funds and by 1996 the reporting

was mandatory for all funds. In its report, each fund lists its individual monthly inflows

and outflows over the previous six months along with answers to various other questions

about the fund operations over the same period.2 The other questions we focus on for this

study concern the loads paid to brokers. Unlike other data sources, these files indicate the

actual dollar value of loads paid by investors. In addition, if a front load is paid, we are

able to determine how much is allocated to the principal underwriter (or management

company) and how much is paid to a captive sales force or unaffiliated broker to


2
 An example of the N-SAR file questionnaire is available at http://www.sec.gov/about/forms/formn-
sar.pdf. The individual data on inflows and outflows is provided in Question 28 (a)-(h). The load data we
use is in Questions 29-38.



                                                    5
distribute the fund. This type of data has not been used before and we exploit it in this

paper to determine whether flows vary by the distribution channel. In particular, we are

able to differentiate between captive and unaffiliated broker channels which have been

aggregated together (see Bergstresser, Chalmers, and Tufano (2004) and O’Neal (2004))

even though the incentives of the two differ.

         Because the N-SAR codes are not linked to CRSP, these two databases are

matched by hand. This initial procedure correctly matches 82% of the funds by name.

The matched database is then subjected to a number of filters. These filters are used to 1)

double-check that the correct match with CRSP is made, 2) ensure at least one year of

trailing data is available for the analysis and 3) remove data subject to entry-error.3 The

final sample contains 1665 funds. It is important to note that our definition of a fund

aggregates across all shareclasses for that fund. In CRSP for this time period, there are




3
  Three filters are applied to the sample: NAV matching, a one-year continuous reporting requirement and
data-entry error filters for both flows and loads. First, NAV numbers reported in the SEC filing are
compared to the NAV reported in CRSP. The N-SAR asks for NAV for two different share classes
(without specifying which classes they report for) and we try to match these NAV numbers with those
reported in CRSP. Matching names by hand and then matching NAV through an automated procedure
leaves us with a sample where we are confident that CRSP and the N-SAR data have been matched exactly.
Unfortunately it may also remove valid data points where the NAVs differ slightly. Second, most of the
analysis requires at least 1 continuous year of historical information. As a result, funds that don’t survive
for longer than one year and funds that miss filing an N-SAR for one period are removed from the sample.
Third, the instructions for filing the N-SAR forms clearly indicate that fund families should report the flow
and load data in thousand’s of dollars. However, comparing fund flows and loads with the size of the fund
provided in the N-SAR it was clear that some funds fail to report this information correctly. To address the
failure of some fund families to correctly scale their N-SAR responses, scaling filters are employed. Three
such filters are employed: (1) the reported monthly redemptions or inflows were on average more than
100% of the total net assets for the entire time period, (2) the reported monthly redemptions or inflows in
one month over the entire time period were more than 200% of total net assets and (3) the flow from any
individual month is more than 50% of TNA. This removes around 5% of the sample. Errors in scaling also
arise with the reporting of dollar loads collected. Dollar values of loads are observed where the calculated
percentage loads is larger percent than the reported maximum load. Again this is a problem of reporting in
the wrong units. To address this issue, we remove an observation if the percent collected in loads divided
by the total dollars of inflows subject to a load is greater than the maximum load. In the case of back loads,
we consider it as a percent of redemptions. The maximum load cutoffs are 8.5%, 6%, and 3% for front
loads, back loads, and redemption fees.



                                                      6
6607 individual funds by shareclasses listed. When aggregating across shareclasses, this

reduces to 2946 separate funds, so we are reporting around 56% of the CRSP sample.

The matched sample of data we are reporting on for most of the sample only includes

funds that are classified as aggressive growth (AG), large-cap growth (LG), and growth

and income (GI). Because we collect data as it is reported, this data does not suffer from

survivorship bias as we capture all the funds that reported after 1996 even if they ceased

to report in later periods. In Tables 6 to 8, we report flows into the entire complex and

hence expand the sample to include all share classes. In this sample with all ICDI

objective categories, there are 3697 separate funds of which 678 are international

(including global bond, global equity and international equity), 442 are specialty funds

(including precious metals, utilities, total return, and specialty funds), and the remaining

912 are bond or money market funds. The average management complex for our sample

has about 15 funds in the complex (the median complex has 9 funds).

3.     Descriptive Analysis

       Table 1 summarizes the filtered dataset of the growth funds with objective

category AG, GI, or LG. The table reports the fund average of new fund purchases,

redemptions, reinvestments and net flows in both thousands of dollars and as a percent of

the fund’s total net assets. Separating net flows into purchases and redemptions is

especially insightful when looking at the decrease in total net assets under management

after 2000. While observing only the net flows could lead the casual observer to interpret

this downward shift in assets under management as an increase in redemptions, in

percentage terms, redemptions actually fell from 2000 to 2001. The fund industry’s

shrinkage over this period was thus not due not to high redemptions, but rather just the




                                              7
market’s decline and lack of reinvestment and new purchases after 2001. While this

point isn’t critical to the rest of our analysis, it does serve to highlight the shortcomings

of net flows relative to separate purchase and redemption information. The ability to

separate these two types of flows is critical to the rest of the analysis.

        Table 2 summarizes payments to brokers. The key result here is that these load

payments bear little resemblance to the maximum loads that are widely reported and used

in academic research. Loads are both negotiable and subject to volume discounts, and

these factors, and not maximum loads, are apparently the major determinants of investors

effective loads.4 What Panel B shows is that, in a comparison across all funds of

maximum loads to the actual load payments we see (first row), the correlations of Front

and Back loads are 22% and 51%, respectively, and in a comparison of load funds only

(second row), the correlations are just 18% and 41%. Results are even weaker for

redemption fees. Because each fund observation mixture of shareclasses we also

calculate these correlations when we condition on a positive load and ensure that over

80% of the fund assets come from either the A shareclass for front loads and the B

shareclass for back loads. Again the correlations are very low, 16% and 30%

respectively. So while maximum loads identify what small investors pay and help

identify shareclasses and no-load funds (see Nanda, Wang and Zheng (2004b), O’Neal

(1999), and Livingston and O’Neal (1998)), they relate weakly to the money actually

being paid.




4
 “Typical breakpoint discounts apply to purchases at $50,000, $100,000, $250,000, $500,000 and $1
million, although some funds provide a breakpoint at $25,000.” Quoted from the March 2003, Joint
SEC/NASD/NYSE Report of Examinations of Broker-Dealers Regarding Discounts on Front-End Sales
Charges on Mutual Funds http://www.sec.gov/news/studies/breakpointrep.htm.


                                                 8
       Panel A of Table 2 shows about 15% of fund inflows are subject to loads, where

the total amount paid is around 39bp in 1996, dropping to 27bp by 2003. The average

effective load for load-fund investments works out to 2.68%, much smaller than the

average maximum front-end load of 4.73% reported concurrently by CRSP. Dollar-

weighting these actual loads paid gives an even lower average load paid.

       The main goal of the paper is to compare investment through three main channels:

captive brokers, unaffiliated brokers and no-load investment. Captive brokers are brokers

representing only one family, whereas unaffiliated brokers in principle represent no

particular family, and no-load investment comes in with no broker. We execute this

comparison by using the NSAR’s decomposition of load payments to sort funds into the

three buckets.

       The NSAR data decompose load income by its destination: captive brokers,

unaffiliated brokers, and the underwriter. We use this decomposition to identify funds

that use only captive brokers, and funds that use only unaffiliated brokers. Many of the

funds in the other category use both, but these are less useful for our purposes because

their redemptions are problematic to interpret. That is, we can estimate from the broker

payments how much of their inflows came from captive vs. unaffiliated brokers, we have

no guidance for estimating the brokerage channel associated with their redemptions.

Payments to the underwriter are essentially payments to the fund family, typically

payments to a distribution affiliate of the management company. These may represent

the fund family’s profit on the sale, but they could also cover other marketing expenses or

cost for underwriting the security. Therefore, for the remainder of the paper we will

define funds as follows. A fund is CAPTIVE if loads are paid to either the affiliated




                                             9
underwriter or captive brokers but not to unaffiliated brokers, UNAFIL if loads are paid

to unaffiliated but not captive brokers, and NOLOAD if no loads are paid at all and no

money is allocated to a broker from the 12b1 fee (listed in Q 42(d) of the N-SAR file). In

the analysis we also report the results for a ONLY12B1 category and an OTHER

category. The ONLY12B1 consists of those funds which do not charge a load but still

pay a broker from their 12b1 payments. The other category is comprised of all other

funds in the sample that could not be allocated to one of the preceding four categories.

While the analysis focuses only on the captive, unaffiliated and no-load channels, the

other two channels are included for completeness. The magnitudes of investment through

these categories are summarized in Table 3.5

         We can relate our database to those of earlier work by replicating (as in O’Neal,

2004) the flow/performance analysis of Ippolito (1992), Sirri and Tufano (1998) and

others. The advantage of our database, relative to the earlier work, is the disaggregation

of monthly flows into purchases and redemptions. We rank funds by their returns for

each ICDI objective category (AG, GI, LG) in the year prior to the flow (Figure 1). The

monthly purchases, redemptions, and reinvestments are divided by beginning of year

total net assets for each month and multiplied by 12 to annualize the percentage change in

flows. These graphs make an important point. While the inflow-performance

relationship is similar to the net flow-performance convexity documented elsewhere in

the literature, the redemption-performance relationship is U-shaped. It might seem from

the earlier studies that investors are reluctant to sell losers, but what we find is that

5
 While the decline in reinvested dollars from 2000 to 2003 looks dramatic, the evidence here is consistent
with evidence reported in the 2005 ICI Fact Book documenting the dollar values of capital gains and
dividends reinvested for equity funds over the same period. Interestingly enough, the percent of capital
gains and dividends that are reinvested actually increases from 2000 to 2001, but dramatic decreases
overall capital gains paid and more subtle decreases in dividends paid account for this significant decrease.


                                                     10
inflows make little distinction between bad and very bad performance, but outflows do

make a significant distinction. Aggregating inflows, redemptions and reinvestments (not

pictured in the graph), the net effect is that flow-performance convexity does not appear

to be as pronounced as in earlier studies.

       Figure 2 shows the redemptions by broker channel. Both the captive and

unaffiliated channels appear to be u-shaped with investors in these channels selling both

the best and worst performers. Of the intermediated channels, unaffiliated has the largest

percentage outflows from poor performers. In comparison, the no-load channel does not

appear u-shaped and redemptions through this channel seem the least sensitive to

performance in the lowest performance category.

4.     Empirical Analysis

The remainder of the paper analyzes the role of intermediaries and how they may alter an

investor’s behavior, measured by the flow-performance relationship. O’Neal (2004) and

Bergstresser, Chalmers, and Tufano (2004) are most closely related to our paper in that

they also consider whether brokers provide a value-added to investors. The real

divergence of our paper with these is our ability to separate brokers by their affiliation,

captive versus unaffiliated, as well as identify the amounts paid to these channels. In

Section 4.A, we discuss how our results compare to those found in these papers. Sections

4.A and 4.B focus on what the investor gets from the intermediary in terms of advice

while sections 4.C and 4.D consider what benefits the mutual fund gets through each

channel.

4.A    Fund Flows and Past Returns




                                             11
        How are fund investments influenced by recent returns? We already know that

net flows generally increase with recent return, and that this relation is convex; our goal

here is to disaggregate the flows to learn about the underlying investment decisions. To

prepare for this, it is worth listing the decisions we are interested in, and considering what

they would look like in the data:

    •   Chasing High Performance: The performance-persistence literature documents

        high future performance of the recent top performers, and the fund-flow literature

        shows these funds attracting net flows. This should be apparent as a positive

        sensitivity of inflows to recent performance in the region of high recent

        performance.

    •   Abandoning the Worst Performers: The performance-persistence literature also

        documents very low future performance of the recent bottom performers.

        However, the fund flow literature does not show these funds losing much net

        flow. To the extent it occurs, it should be apparent as a negative sensitivity of

        redemptions in the region of low recent performance.

    •   Rebalancing out of Winner Funds: Presumably, high performance moves

        investors portfolio weights out of line, requiring the investors to redeem some to

        restore proper diversification. Such rebalancing would be apparent as a positive

        sensitivity of redemptions in the region of high recent performance.

    •   Rebalancing into Loser Funds: Similarly, low performance requires investors to

        invest more to restore proper diversification. This effect would be apparent as a

        negative sensitivity of inflows in the region of low recent performance.




                                             12
Our goal is to determine how the significance of these motives varies across the forms of

intermediation.

         We begin with a simple test design. We explain percentage inflows and

redemptions through each channel with ranked one-year past return. To avoid a potential

bias, the calculation of percentage inflows is slightly modified from the usual calculation

in the literature: we divide flows by fund size at the beginning, rather than the end, of the

one-year return. If we used the end size, then a zero relation between dollar flows and

performance would show up as a negative relation between percentage flows and

performance (that is, good performance shrinks the percentage by increasing the

denominator).6 In Tables 5-8, when we relate a fund’s percentage flows to quantities

other than past performance, this concern does not apply so we use the standard

calculation with the end size.

         The details are as follows. For each month, we calculate each fund’s trailing one-

year return, and within each fund category (AG, GI and LG) we convert returns to

percentile ranks Rkret from 0 to 1 (1 being the best). Since these flow decisions apply to

different regions of past returns, we can allow for different slopes in the different regions.

Thus, we separate Rkret into three pieces: Rkretlo, which is min{Rkret, 0.2}, Rkrethi,

which is max{Rkret-0.8, 0}, and Rkretmed=Rkret-Rkretlo-Rkrethi. With this

specification, abandoning the worst funds appears as a negative coefficient of

redemptions on Rkretlo, and rebalancing out of winner funds appears as a positive

coefficient on Rkrethi. We also include indicator variables for the channels. For each

month and year, the cross section of percentage flows is regressed on the dummies, Rkret


6
  In earlier versions of this paper (available on request) we used the end size, and slopes were indeed biased
in this fashion.


                                                     13
interacted with them, and log of total net assets as a control. We average the monthly

coefficients and calculate Newey-West standard errors with 5 lags (see Pontiff (1996)).

Results are in Table 4, with Redemptions, Inflows and Net Flows in separate columns.7

        In Table 4 we see that redemptions through the captive and no-load channels have

a positive and statistically significant coefficient and the unaffiliated channel has a

positive and marginally significant coefficient on Rkrethi, consistent with the notion of

rebalancing out of winner funds. Looking at the Rkretlo coefficient for redemptions, we

are able to examine our prediction that the Captive and No-Load channels are less likely

to be sensitive to poor performance. The No-Load channel shows significantly less

evidence of abandoning the worst performers than do the other channels. However, the

Captive channel, despite having a smaller coefficient (in absolute terms) than the

Unaffiliated channel, is not statistically different.

        The weak response to bad returns of no-load investors, compared to investors

with either kind of broker, is telling. Investors in the worst funds may be “sleeping” or

otherwise insensitive to reinvestment opportunities (e.g. Christoffersen and Musto, 2002,

Berk and Xu, 2004), possibly due to psychological difficulties with bad outcomes of

personal decisions (e.g. Goetzmann and Peles, 1996). Such an investor benefits from

having someone to wake him up, or otherwise shake his complacency, and it appears that

brokers do just that. So value brokers add may be at least as much on the way out as on

the way in.

        Before moving on to inflows, it is worth considering the evidence for the

disposition effect of Shefrin and Statman (1985), generally associated with stock, rather


7
 It is important to note that Net Flows are not exactly Inflows minus Redemptions, due to reinvested
dividends.


                                                    14
than mutual fund, investing. It is the observation that investors tend to sell winners

quickly but hang on to losers. This would manifest, in our data, as high positive

sensitivity of redemptions to high returns, and low sensitivity of redemptions to low

returns. Looking across the channels, investments through unaffiliated brokers show the

least of this behavior, and no-load investments show the most. So while it might seem

from net flows that the disposition effect does not apply, it is actually a good description

of the behavior of no-load investors, particularly when contrasted with the unaffiliated

brokers.

        The questions for inflows are whether they show positive sensitivity in the region

of good returns associated with chasing high performance, and negative sensitivity in the

region of bad returns associated with rebalancing into loser funds. We see strong

evidence for of chasing high performance but no evidence for rebalancing into loser

funds. All channels show strong positive sensitivity in the high-return region, and none

show significant slopes in either direction in the low-return region. Of course, shifting

into poor performers for rebalancing would have the opposite effect of avoiding poor

performers, so it is not a surprise that the net effect is a wash.

4.B     Fund Flows and Future Returns

        The results of Table 4 beg the question, how does the investment decision turn

out? Does performance increase with inflows and decrease with redemptions, and in

particular, does this future performance vary with the form of intermediation? We

address this question by regressing future six-month excess return, i.e. return in excess of

the average for the fund’s category, on percentage flows (calculated from end, rather than

beginning, sizes) interacted with the dummies for intermediation channel. Again,




                                               15
regressions are cross-sectional and monthly, and standard errors are adjusted for

autocorrelation. Our sample period is somewhat short for inferences about expected

returns, but it does contain both up and down markets. The results are in Table 5.

       To interpret Table 5, bear in mind that a positive coefficient in the Net Flows and

Inflows columns indicates good news for investors, i.e. better future performance if more

money came in, and a positive coefficient in the Redemptions column indicates bad

news, i.e. better future performance if more money left. So what the first row of Panel B

tells us is that inflows through captive brokers pan out significantly better than those

through unaffiliated brokers, whereas the reverse holds for redemptions. For net flows,

only no-load flows relate significantly to future returns, but the point estimates are similar

across channels, with no significant difference. So while our short time series can

discriminate only so much, we do get evidence that among brokers, captive brokers add

relatively more value on the way in and unaffiliated brokers add more value on the way

out, but only net flows through the no-load channel predicts future performance.



4.C Fund Family Economics

       From the perspective of a fund’s family, an inflow to a fund is less valuable if it

would otherwise have gone to another fund in the family, and a redemption is less costly

if it is reinvested at another fund in the family. While SEC filings do not directly show

money moving from one fund to another, we can still indirectly gauge these flows by

relating the inflows and redemptions of a fund to the simultaneous reverse flows of other

funds in the same family.




                                             16
       In this section, we must distinguish between a fund’s predicted and excess inflows

and redemptions. The predicted flows are the flows we would expect, given the fund’s

operating history but ignoring family effects. For predicted flow, we use the prediction

of the piecewise regression model of Table 4. The residual from this model, capturing

the family effects is the excess flow. For each fund we also calculate the predicted and

excess flows of the rest of the family from the same numbers: the predicted and excess

percentage flows of all other funds in the family, converted to dollars (i.e. each multiplied

by the fund’s total net assets), summed and divided by the funds’ aggregate total net

assets. Note that this includes all funds in the family, not just equity funds. With these

definitions, we can address three likely family-level fund-flow concerns:

   •   Recapture: A family recaptures flow if an expected redemption from one fund

       becomes an unexpected inflow elsewhere within the family. So excess family

       inflows, excluding the fund, should be increasing in predicted fund redemptions.

   •   Rebalancing out of winners: Rebalancing is like recapture, except that the flows

       would come from winners, and would go to a different asset class. So excess

       non-equity-fund family inflows should be increasing in predicted fund

       redemptions, particularly among recent-winner funds.

   •   Cannibalization: A family experiences cannibalization if extra inflow to one fund

       comes out of inflow to the rest of the family. So excess family inflows, excluding

       the fund, should be decreasing in excess fund inflows.

As before, the goal is to understand how these concerns vary across the forms of mutual-

fund intermediation.




                                             17
        We first address recapture, and the regression model is adapted from Table 5. For

every month, we run a cross-sectional regression where the dependent variable is the

percentage excess inflow into a family, omitting one fund, and the independent variables

are the channel dummies, alone and interacted with the predicted redemptions of the

omitted fund. We also include log fund size and log complex size as controls, and we

drop funds with only 0 or 1 other funds in the family. In addition to the regression using

all funds, we run two more, one with only funds with above-median performance in the

recent year, and one with funds below the median. Again, monthly coefficients are

averaged and standard errors are adjusted for autocorrelation. Results are in Table 6.

        Only one channel exhibits recapture, and it is the one we would expect: the

captive channel. Thus, an in-house brokerage force retains assets that would otherwise

leave, a benefit for the family. This is stronger if the fund being left did better, but it is

significant in either case.

        To address rebalancing we run the same model, the only difference being that the

dependent variable is excess flows into only the non-equity funds in the family. The

results, in Table 7, again show only captive brokers succeeding in this regard.

Comparing the high-performer and low-performer results for captive funds, we see a

higher slope among higher performers, where we would expect more rebalancing, but the

difference is not statistically significant.

        Intuitively, the benefits of recapturing redemptions come at the cost of

cannibalizing inflows. Unlike unaffiliated brokers, captive brokers are poorly situated to

bring in flows that would otherwise go to other families, so increasing one fund’s flows is

more likely to come at the expense of other funds in the family. We test for cannibalizing




                                               18
by returning to the dependent variable of Table 6, excess inflows to the rest of the family,

and changing the independent variable from predicted redemptions to excess inflows. If

a fund’s positive surprise inflows are the rest of the family’s negative surprise, we will

see a negative loading. We also test for cannibalization by changing the independent

variable to ranked return, Rkret. The more a fund boosts its inflows by increasing its

share of the dollars flowing to the family, rather than by increasing the dollars coming

into the family, the more negative this should enter. Results are in Table 8.

           We see from both perspectives that captive funds suffer from cannibalization

significantly more than the other funds. So by frustrating flows between their family and

other families, the families opting for in-house brokers both succeed at retaining assets

but fail at attracting them from elsewhere.




4.D. Broker Incentives

           Finally, we can explore the idea, popular in the press recently, that unaffiliated

brokers favor funds that pay them more. For example, PIMCO’s distributor (PA

Distributor) recently settled with California’s Attorney General for $9 million. The legal

suit alleged that PIMCO and the distributor engaged in a shelf-space arrangement

whereby broker-dealers were required to “tout PIMCO mutual funds, via placement on

intranet web sites or ‘preferred’ or ‘recommended’ lists.”8 Similarly, Eliot Spitzer was




8
    http://ag.ca.gov/newsalerts/2004/04-105.htm


                                                  19
quoted as saying “fees, fees, fees are the next issue” where there are growing concerns

that a broker’s objective advice to a client is compromised as they are paid more.9

        We have limited ability to pursue this theory, because we do not see how much

any particular broker is paid for one fund vs. another. But we can see which funds pay

unaffiliated brokers relatively more, and which pay captive brokers relatively more, and

we can see whether this relates to the level of flows. The dependent variable is

percentage flows to the fund and the new explanatory variables are HICAP, which is 1

for captive-broker funds paying more than the mean captive-broker amount to their

brokers, and HIUNAF, which is the analogous statistic for unaffiliated-broker funds. The

results are separated by performance with Panel A and Panel B relating to the low

performers (bottom 50% of ranked returns) and Panel C and Panel D relating to high

performers (top 50% of ranked returns). The results are in Table 9, with captive brokers

in Panels A and C and unaffiliated brokers in Panels B and D.

        For the low performers in Panels A and B both HICAP and HIUNAF do not enter

for net flows for any of the specifications, though both are associated with lower

redemptions and lower inflows. This seems odd, but it would result naturally from these

funds having smaller average investments, which would ramp up loads (due to the

vanishing volume discounts) and shrink investment amounts. The interaction between

HICAP and Rkret does not enter in for the purchases or redemptions in either the high or

low performers. For the unaffiliated brokers in the low performance funds, higher

compensation is associated with lower sensitivity to poor performance, as manifested by

the positive coefficient on RkRet*HIUNAF, but the result is not statistically significant.


9
 http://www.smh.com.au/articles/2003/12/04/1070351723857.html. And see also
http://www.sec.gov/news/press/2003-159.htm for the Morgan Stanley case.


                                               20
Also, the sensitivity of inflows to positive performance in the high performance category

is increased when the unaffiliated broker’s commission is higher and the result is

statistically significant. The contrast between the captive channel, where one would not

expect payment for favoritism, and the unaffiliated channel, where one might, is at least

suggestive of a purchase of extra effort, both to bring flows into good performers and to

keep them in poor performers.



5.      Conclusion

        Money flows in and out of mutual funds through different channels, and these

channels represent different amounts and types of intermediation and advice. In this

paper we explore the economics of the three main channels – captive brokers, unaffiliated

brokers and the no-load, direct channel – to see what drives these flows, and what they

accomplish for consumers. Our main findings are:

     1) The loads consumers pay have little relation to the maximum loads usually

        referenced, and they are smaller for captive brokers than for unaffiliated brokers.

     2) Redemptions by no-load investors are significantly less sensitive to bad

        performance than are redemptions by investors with brokers.

     3) Captive brokers add more value than Unaffiliated brokers on the way into funds,

        but the reverse is true on the way out.

     4) Fund families benefit from captive brokerage through recapture of redemptions

     5) However, they also suffer through cannibalization of inflows.

We also find evidence that higher payments to unaffiliated brokers buys some amount of

favorable treatment in attracting inflows.




                                             21
       Rational decision theory helps us understand consumers’ mutual fund decisions

only to a point, and that point is determined by both the sophistication of investors who

go it alone, and the rationality and incentives of the professionals that investors turn to for

help. Since the literature both expects (e.g. Berk and Green, 2004) and finds little

predictability in fund returns, it is perhaps not a surprise that the main effect of brokerage

advice is not on the consumers getting the advice but on the families getting their fees.

We find that families opting for captive brokers experience the associated tradeoff: more

retention of funds, but less acquisition. The question of why some families prefer this

tradeoff but others do not, we leave for future research.




                                              22
References

Alexander, G. J., J. D. Jones, and P. J. Nigro, 1998, “Mutual Fund Shareholders:
Characteristics, Investor Knowledge and Sources of Information,” Financial Services
Review 7, 301-316.

Barber, Brad M., Terrance Odean, Lu Zheng, 2004, Out of Sight, Out of Mind: The
Effects of Expenses on Mutual Fund Flows, forthcoming in Journal of Business.

Berk, Jonathan B., and Richard C. Green, 2004, Mutual fund flows and performance in
rational markets, Journal of Political Economy 112, 1269-1295.

Berk, Jonathan B., and Jing Xu, 2004, Persistence and fund flows of the worst
performing mutual funds, working paper, University of California, Berkeley.

Bergstresser, Daniel, John Chalmers, Peter Tufano, 2004, Assessing the Costs and
Benefits of Brokers: A Preliminary Analysis of the Mutual Fund Industry, Unpublished
Manuscript, Harvard.

Brown, Keith C., W. V. Harlow and Laura T. Starks, 1996, Of Tournaments and
Temptations: An Analysis of Managerial Incentives in the Mutual Fund Industry, The
Journal of Finance, 51, 85-110.

Brown, Stephen J., and William N. Goetzmann, 1995, Performance persistence, Journal
of Finance 50, 679-698.

Capon, N., G. Fitzsimmons and R. Prince, 1996, “An Individual Level Analysis of the
Mutual Fund Investment Decision,” Journal of Financial Services Research 10, 59-82.

Carhart, Mark M., 1997, On persistence in mutual fund performance, Journal of Finance
52, 57-82.

Carhart, Mark M., Jennifer N. Carpenter, Anthony W. Lynch and David K. Musto, 2002,
Mutual Fund Survivorship, Review of Financial Studies, 15, 1439-1463.

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

Christoffersen, Susan, and David Musto, 2002, Demand Curves and the Pricing of Money
Management, Review of Financial Studies, 15, 1499-1524.

Goetzmann, William N., and Nadav Peles, Cognitive Dissonance and Mutual Fund
Investors, The Journal of Financial Research 20 (Summer 1997), 145-158.

Gruber, Martin J., 1996, Another puzzle: The growth in actively managed mutual funds,
Journal of Finance, 51, 783-810.



                                           23
Hendricks, Darryll, Jayendu Patel, and Richard Zeckhauser, 1993, Hot hands in mutual
funds: Short-run persistence of relative performance, 1974-1988, Journal of Finance 48,
93-130.

Ippolito, Richard, 1992, Consumer Reaction to Measures of Poor Quality: Evidence From
the Mutual Fund Industry, Journal of Law and Economics, 35, 45-70.

Johnson, Woodrow, 2004, Predictable Investment Horizons and Wealth Transfers among
Mutual Fund Shareholders, Journal of Finance 59, 1979-2012.

Livingston, Miles, and Edward O’Neal, 1998, “The Cost of Mutual Fund Distribution
Fees”, Journal of Financial Research, 21(2), p. 2015-218.

Nanda, Vikram, Z. Jay Wang and Lu Zheng, 2004a, Family Values and the Star
Phenomenon, Review of Financial Studies, 17, 667-698.

Nanda, Vikram, Z. Jay Wang and Lu Zheng, 2004b, The ABCs of Mutual Funds: A
Natural Experiment on Fund Flows and Performance, Working Paper, U. Michigan.

Odean, Terry, 1998, Are investors reluctant to realize their losses? Journal of Finance,
53(5), 1775-1798.

O’Neal, Edward, 1999, Mutual Fund Share Classes and Broker Incentives Financial
Analysts Journal, 55(5), 76-87.

O’Neal, Edward, 2004, Purchase and Redemption Patterns of US Equity Mutual Funds,
Financial Management, Spring, 63-90.

Pontiff, J., 1996, Costly arbitrage: Evidence from closed-end funds, Quarterly Journal of
Economics 111, 1135-1152.

Reuter, Jonathan, 2005, Are IPO Allocations For Sale? Evidence from Mutual Funds,
Journal of Finance, forthcoming.

Shefrin, H and M. Statman, 1985, The disposition to sell winners too early and ride losers
too long: theory and evidence, Journal of Finance 40, 777-792.

Sirri, Erik R., and Peter Tufano, 1998, Costly search and mutual funds flows, Journal of
Finance 53, 1589-1622.

Zheng, Lu, 1999, Is money smart? A study of mutual fund investors’ fund
selection ability, Journal of Finance 54, 901-933.




                                            24
                                                     Performance and Flows

                      120
                                                      Redemptions             Inflows          Net flows
                      100
Annualized Flow (%)




                      80

                      60

                      40

                      20

                        0
                            1        2           3          4          5          6           7          8          9           10
                      -20
                                                                Performance Decile

                            Figure 1. Percentage Flows and Current Performance. This graph plots
                            annualized monthly new inflows, redemptions, and net flows as a percent of total net assets (from
                            the previous year) by the performance decile. Here decile 1 is the lowest performance decile and
                            decile 10 is the highest. Performance is evaluated over the year prior to the monthly flows and
                            ranked by investment objective category.




                                                                       25
                                                              Redemptions By Channel

                             70

                             65
                                                                  Captive            Unafil        Noload
Annualized Redemptions (%)




                             60

                             55

                             50

                             45

                             40

                             35

                             30

                             25

                             20
                                  1           2           3          4           5            6         7          8           9         10
                                                                         Performance Decile
                                      Figure 2. Percentage Redemptions by Investment Channel. This graph plots
                                      annualized monthly redemptions as a percent of total net assets (from the previous year) by the
                                      performance decile. Here decile 1 is the lowest performance decile and decile 10 is the highest.
                                      Performance is evaluated over the year prior to the monthly flows and ranked by investment
                                      objective category. Redemptions are plotted for the captive, unaffiliated, and no-load channels.




                                                                                 26
                                  Table 1. Descriptive Statistics
This table provides descriptive statistics of our matched sample of CRSP and N-SAR files from 1996 to 2003.
Between 1996-2003, there are 98,352 fund/month observations. These only include equity funds classified as
either AG, LG, or GI according to the ICDI objective categories in CRSP where one fund observation groups all
shareclasses together. The purchases, redemptions, and reinvestments are based on information from Question 28
(a) – (h) in the NSAR file. Panel A provides the annual summary statistics while Panel B provides the monthly
statistics for the same time horizon. All values are fund averages.

Variables                    Units       1996 1997 1998 1999 2000 2001 2002 2003
Purchases                   ($ '000)    26906 31092 32801 39366 47596 31514 27360 25327
Redemptions                 ($ '000)    20264 24775 28252 34339 38917 27352 25556 20718
Reinvestments               ($ '000)      6352 8348 5519 8492 10323 1669           578    624
Net flows                   ($ '000)    13027 14680 10193 13534 19050 6785 2395 5266
Purchases                   % TNA         4.14  4.60   4.53   4.33   4.50   4.20  3.93   3.14
Redemptions                 % TNA         2.96  3.33   3.93   4.15   3.94   3.56  3.79   2.91
Reinvestments               % TNA         0.61  0.73   0.47   0.55   0.75   0.15  0.06   0.05
Net flows                   % TNA         1.83  2.05   1.12   0.75   1.32   0.84  0.20   0.29
TNA                        $ millions    801.5 987.8 1083.4 1204.3 1442.4 1098.4 941.6 1015.6
Expenses                       %          1.29  1.26   1.26   1.27   1.26   1.27  1.33   1.39
Month/Fund Obs                           7625 8828 10011 11337 12106 12392 12346 11524




                                                  27
                                  Table 2. Summary statistics on actual loads paid

This table provides descriptive statistics of our matched sample of CRSP and N-SAR files from 1996 to 2003. Between
1996-2003, there are 98,352 fund/month observations. These only include equity funds classified as either AG, LG, or GI
according to the ICDI objective categories in CRSP where one fund observation groups all shareclasses together. The load
information and its allocation by sales force is based on information from Question 28-38 in the NSAR file. Panel A
provides the annual summary statistics while Panel B provides correlation between the actual loads collected that are
reported on the N-SAR files and the maximum load charged which are reported in CRSP. Percent of inflows subject to a
load is the percent of total dollars arriving into a fund subject to a load (Q28(h)) divided by the total dollars arriving into
the fund (Q28(g)). The percent of inflows collected in front loads is the total dollars collected in front loads Q30(a) divided
by Q28(g). The percent of inflows paid to the underwriter is the total dollars paid to an underwriter (Q31a) divided by the
total inflows (Q28(g)). The percent of inflows paid to the unaffiliated broker is Q32 divided by Q28(g) while the percent of
inflows to a captive sales force is Q33 divided by Q28(g). The percent of outflows subject to back loads is Q35 divided by
the total outflows for the six month period given as part of the response to Q28(g). The percent received by brokers in a
specific channel for a fund is the percent of dollars received in either the captive (Q33) or unaffiliated channel (Q32)
divided by the dollars coming to that fund which are subject to a load (Q28(h)). The percent paid by individuals is the total
front load dollars collected (Q30(a)) divided by the total inflows subject to a load, Q28(h). The proportion of 12b1 fees
paid to brokers is reported in Q42(d). The conditional correlations condition on the front load, back load, and redemption
fee being positive as reported in CRSP. The last two rows condition on the fund having over 80% of its assets in the A
shareclass and over 80% its assets in the B shareclass. Note that CRSP refers to a deferred contingent sales charge or back
load as a rear load. CRSP also calls the redemption fee imposed on investors who trade within a short time frame as a
deferred load.

Panel A: Load Information                                1996 1997            1998     1999     2000     2001      2002    2003
Percent of inflows subject to loads                     16.51 15.47          14.42    13.33    13.51    14.14     15.93    16.12
Percent of inflows collected in front loads               0.39    0.37         0.30     0.26     0.25     0.24     0.26     0.27
   Allocated to underwriter (% inflow)                    0.12    0.11         0.08    0.07      0.06     0.06     0.07     0.07
   Allocated to captive salesforce (% inflow)             0.11    0.11         0.07     0.06     0.05     0.04      0.06    0.07
   Allocated to unaffiliated broker (% inflow)            0.20    0.20         0.20     0.17     0.15     0.14      0.18    0.19
Percent of outflows collected in back loads               0.07    0.07         0.10     0.10     0.10     0.11     0.12     0.12
Percent of front loads allocated to
   Underwriter (% front load collected)                 27.30 29.09          25.27    20.70    22.27    22.35     23.31    19.66
   Captive Salesforce (% front load collected)          21.74 21.50          20.83    22.81    21.17    20.68     24.28    19.36
   Unaffiliated broker (% front load collected)         50.15 49.51          55.32    58.65    59.69    59.75     60.32    62.82
Average load received by brokers
   Captive                                                1.04    0.93        0.92     1.07      0.97    0.99      1.19     1.42
   Unaffiliated                                           2.41    2.32        3.81     2.41      2.40    2.26      3.31     2.78
Average load paid by investors
   Captive                                                2.67    2.54        2.10     2.13      2.12    2.17      2.33     2.62
   Unaffiliated                                           2.88    2.86        2.72     2.64      2.64    2.62      2.60     2.65
Portion 12b1 paid to brokers
   Captive                                              26.46 27.32          19.93    16.58    18.44    27.90     27.07    15.77
   Unaffiliated                                         31.67 35.47          34.83    42.19    39.41    34.52     31.96    27.50
   Only 12b1                                            58.06 70.58          75.15    80.11    83.38    78.38     78.79    78.18
Panel B: Correlation between NSAR Actual Loads and CRSP Maximum Loads
                                                         Front Load                   Back Load                  Redemption Fee
Unconditional                                              0.2195                      0.5126                        0.3047
Conditional on Positive Load                               0.1844                      0.4071                       -0.2543
Conditional on Positive Load and >80% in A Shareclass      0.1636
Conditional on Positive Load and >80% in B Shareclass                                   0.2987




                                                               28
                                Table 3. Total flows by distribution channel

This table provides descriptive statistics of our matched sample of CRSP and N-SAR files from 1996 to 2003. These only
include equity funds classified as either AG, LG, or GI according to the ICDI objective categories in CRSP where one fund
observation groups all shareclasses together. NOLOAD represents funds that are not subject to a load and do not pay a
12b1 fee. CAPTIVE is defined as a fund paying a front load which is allocated to either the underwriter or captive sales
force but none is allocated to an unaffiliated broker. UNAFFILIATED is one if a front load is paid and is allocated to the
unaffiliated broker but not to the captive sales force. ONLY12b1 is defined as a fund charging no load but there is still an
allocation to the broker from 12b1 fees. OTHER funds include all remaining funds. SOME CAPTIVE implies some of the
load was allocated to a captive salesforce. SOME UNAFFILIATED is defined as a fund allocating some of the load to an
unaffiliated salesforce.

                                            1996      1997     1998       1999     2000      2001      2002      2003
Total Dollars Redeemed ($bil)               152.2     214.6    279.5      382.0    467.6     337.4     312.2     236.9
 Noload (%)                                  26.9      26.9     34.3       33.6     35.3      38.6      34.7      33.8
 Only Captive (%)                            19.9      19.2     11.2        9.1     10.2      14.0      13.6      10.2
 Only Unaffiliated (%)                       21.4      25.5     33.9       38.0     36.1      33.5      37.8      37.1
 Only 12b1 (%)                                2.7       2.9      2.5        3.4      3.5       3.4       3.7       4.0
 Other (%)                                   29.0      25.5     18.2       15.9     14.9      10.5      10.2      14.9
 Some Captive (%)                            26.7      23.7     14.8       12.5     12.4      16.6      16.6      15.9
 Some Unaffiliated (%)                       28.2      30.0     37.5       41.3     38.3      36.0      40.7      42.8

Total Dollars Purchased ($bil)              202.1     269.3    324.6      437.9    571.9     388.8     334.3     289.6
 Noload (%)                                  23.4      24.5     31.9       29.8     34.8      36.3      31.9      30.4
 Only Captive (%)                            17.6      17.4     10.4        7.3      7.7      12.5      11.9       8.4
 Only Unaffiliated (%)                       24.8      29.5     36.7       38.8     37.4      37.0      38.1      37.7
 Only 12b1 (%)                                3.2       2.5      2.1        3.4      3.6       2.9       4.1       4.3
 Other (%)                                   30.9      26.2     18.8       20.7     16.6      11.3      14.0      19.1
 Some Captive (%)                            23.5      21.2     13.5       10.2     10.1      14.7      14.2      14.0
 Some Unaffiliated (%)                       30.8      33.4     39.8       41.7     39.9      39.1      40.3      43.4

Total Dollars Reinvested ($bil)              47.7      72.3        54.6    94.5    124.0      20.6       7.1       7.1
 Noload (%)                                  22.8      29.1        37.9    32.1     34.4      26.8      39.7      37.6
 Only Captive (%)                            27.3      19.6         9.8     8.5      9.9      22.7      14.5       5.6
 Only Unaffiliated (%)                       21.4      27.3        41.0    46.4     40.7      40.3      35.7      40.9
 Only 12b1 (%)                                0.8       1.8         1.1     1.9      2.3       3.2       1.9       2.4
 Other (%)                                   27.7      22.3        10.3    11.2     12.7       7.0       8.3      13.5
 Some Captive (%)                            30.7      21.7        12.2    11.3     12.0      24.0      15.4       9.3
 Some Unaffiliated (%)                       24.8      29.4        43.4    49.2     42.8      41.6      36.6      44.5




                                                              29
                                         Table 4. Convexity in fund flows
This table provides regressions of percent of TNA in redemptions, inflows, and net flows on ranked returns of the fund
allowing for low, medium and high performance funds. These only include equity funds classified as either AG, LG, or GI
according to the ICDI objective categories in CRSP where one fund observation groups all shareclasses together.
REDEMPTIONS are the dollar value of redemptions leaving a funds in a given month divided by the total net assets of the
fund 12 months earlier. INFLOWS are the total dollar value of new money coming into a fund in a given month divided by
the total net assets of the fund 12 months earlier. NET FLOWS are the dollar inflows plus reinvestments less redemptions
in a given month for a fund divided by its total net assets 12 months earlier. NOLOAD represents funds that are not subject
to a load and do not pay a 12b1 fee. CAPTIVE is defined as a fund paying a front load which is allocated to either the
underwriter or captive sales force but none is allocated to an unaffiliated broker. UNAFFILIATED is one if a front load is
paid and is allocated to the unaffiliated broker but not to the captive sales force. ONLY12b1 is defined as a fund charging
no load but there is still an allocation to the broker from 12b1 fees. OTHER funds include all remaining funds. RKRETLO
is the min(RKRET, 0.2) where RKRET is the ranked return of the fund over the past 12 months. Ranking is done for every
month/year and compares all the funds in the sample for that month ranking them from 0 to 1. RKRETHI is max(RKRET-
0.8, 0) and RKRETMED is the min(RKRET-RKRETLO,0.6). To adjust for autocorrelation in the panel, we repeat this
regression separately for each month and year between 1996-2003. We report the estimated intercept from regressing the
individual coefficients on a constant and report the t-statistic of this regression using Newey-West standard errors with 5
lags.

Panel A: Convexity in flows by fund
                                                        Redemptions                Inflows                Net flows
                                                      Coef      t-stat        Coef       t-stat       Coef        t-stat
Rkretlo*Captive                                      -63.43     -3.85        -21.54      -1.25        52.69        4.31
Rkretlo*Unafil                                       -84.16     -2.56        -30.85      -1.13        53.37        3.41
Rkretlo*Noload                                       -14.95     -1.31         12.87       0.93        34.63        3.13
Rkretlo*12b1Only                                     -75.12     -2.81         34.99       0.98       120.46        3.75
Rkretlo*Other                                        -66.04     -3.83        -12.47      -0.75        63.93        3.44
Rkretmed*Captive                                       8.12      1.35         57.31       5.82        51.70        5.96
Rkretmed*Unafil                                        5.02      0.79         49.38       4.10        48.42        6.36
Rkretmed*Noload                                        4.76      1.22         50.70       8.07        48.30        7.26
Rkretmed*12b1Only                                     10.02      1.66         51.40       5.79        41.47        4.86
Rkretmed*Other                                        15.65      3.38         54.70       9.45        41.57        7.14
Rkrethi*Captive                                      122.80      3.40        283.77       4.36       180.95        3.30
Rkrethi*Unafil                                        48.61      1.71        206.14       4.56       174.42        5.30
Rkrethi*Noload                                       103.47      6.71        246.02       8.55       151.49        6.22
Rkrethi*12b1Only                                    3052.24      1.12       2547.61       1.23      -490.47      -0.74
Rkrethi*Other                                         66.47      1.90        269.55       5.29       211.37        8.25
Log TNA                                                0.82      2.76          0.53       2.31        -0.11      -0.37
Captive                                               36.60      9.90         23.57       5.94       -11.64      -3.24
Unafil                                                49.17      7.72         38.97       5.15        -7.49      -2.39
Noload                                                33.00     14.87         22.27      10.09        -8.66       -4.50
12b1Only                                              51.03      9.33         28.73       6.10       -20.68      -5.60
Other                                                 39.19     12.64         29.08       9.35        -8.51      -4.01




                                                             30
                            Table 4. Convexity in Flows (Continued)


Panel B: Tests
Differences Across Channel (Low Returns)
 Rkretlo*Captive - Rkretlo*Unafil = 0         20.73      0.53     9.30      0.30    -0.68    -0.04
 Rkretlo*Captive - Rkretlo*Noload = 0        -48.49     -2.08    -34.42    -1.30    18.05     1.33
 Rkretlo*Captive - Rkretlo*12b1Only =0        11.69      0.44    -56.53    -1.36   -67.77    -2.11
 Rkretlo*Unafil -Rkretlo*Noload=0            -69.21     -2.06    -43.72    -1.59    18.73     1.39
 Rkretlo*Unafil - Rkretlo*12b1Only = 0        -9.04     -0.21    -65.84    -1.32   -67.09    -2.56
 Rkretlo*Noload - Rkretlo*12b1Only =0         60.18      2.08    -22.12    -0.63   -85.82    -3.24
Differences Across Channel (High Returns)
 Rkrethi*Captive - Rkrethi*Unafil = 0         74.19      1.38     77.63     1.05    6.53     0.12
 Rkrethi*Captive - Rkrethi*Noload = 0         19.34      0.41     37.75     0.53    29.47    0.53
 Rkrethi*Captive - Rkrethi*12b1Only =0      -2929.43    -1.08   -2263.84   -1.09   671.43    1.02
 Rkrethi*Unafil -Rkrethi*Noload=0            -54.86     -1.87    -39.88    -0.66    22.94    0.47
 Rkrethi*Unafil - Rkrethi*12b1Only = 0      -3003.63    -1.10   -2341.47   -1.12   664.90    1.01
 Rkrethi*Noload - Rkrethi*12b1Only =0       -2948.77    -1.08   -2301.59   -1.12   641.96    0.96
Convexity Captive
 Rkretmed*Captive-Rkretlo*Captive=0         71.55       3.48    78.86      3.09     -0.99    -0.06
 Rkrethi*Captive-Rkretmed*Captive=0         114.68      2.85    226.46     3.32    129.26     2.30
Convexity Unafil
 Rkretmed*Unaf-Rkretlo*Unaf=0                89.18      2.52    80.22      2.53     -4.94    -0.29
 Rkrethi*Unaf-Rkretmed*Unaf=0                43.60      1.37    156.76     2.92    126.00     3.30
Convexity Noload
 Rkretmed*Noload-Rkretlo*Noload=0            19.71      1.45     37.83     2.41     13.66    1.15
 Rkrethi*Noload-Rkretmed*Noload=0            98.71      7.33    195.32     7.20    103.19    4.30
Convexity 12b1Only
 Rkretmed*12b1Only - Rkretlo*12b1Only=0      85.14      2.94     16.41     0.42     -78.99   -2.26
 Rkrethi*12b1Only-Rkretmed*12b1Only=0       3042.22     1.12    2496.22    1.21    -531.94   -0.80




                                                       31
                Table 5. Future returns as a function of flows and distribution channel
This table provides regressions of future six month excess returns on today’s redemptions, purchases, and net flows into a
fund. These only include equity funds classified as either AG, LG, or GI according to the ICDI objective categories in
CRSP where one fund observation groups all shareclasses together. REDEMPTIONS are the dollar value of redemptions
leaving a fund in a given month divided by the total net assets of the fund. INFLOWS are the total dollar value of new
money coming into a fund in a given month divided by the total net assets of the fund. NET FLOWS are the dollar inflows
plus reinvestments less redemptions in a given month for a fund divided by its total net assets. NOLOAD represents funds
that are not subject to a load and do not pay a 12b1 fee. CAPTIVE is defined as a fund paying a front load which is
allocated to either the underwriter or captive sales force but none is allocated to an unaffiliated broker. UNAFFILIATED is
one if a front load is paid and is allocated to the unaffiliated broker but not to the captive sales force. ONLY12b1 is defined
as a fund charging no load but there is still an allocation to the broker from 12b1 fees. OTHER funds include all remaining
funds. XSRET6M is the six month excess return for the six months after the current period where a fund’s monthly excess
return is the difference between the fund’s return and the average return of all funds in that month/year. To adjust for
autocorrelation in the panel, we repeat this regression separately for each month and year between 1995-2003. We report
the estimated intercept from regressing the individual coefficients on a constant and report the t-statistic of this regression
using Newey-West standard errors with 5 lags.

Panel A: XS returns 6 mos ahead on the lagged redemption and purchase decision of investors
DepVar = XS returns 6 mos ahead            FLOWS = %Redemptions        FLOWS=%Purchases               FLOWS=%Netflows
                                              Coef         t-stat       Coef           t-stat          Coef     t-stat
FLOWS*Captive                                 3.65E-06          0.32    1.43E-05            1.07       1.59E-05      0.85
FLOWS*Unafil                                 -2.30E-05         -2.51   -5.23E-06           -0.37       1.48E-05      1.32
FLOWS*Noload                                 -7.01E-06         -0.92    1.48E-05            1.33       1.91E-05      2.14
FLOWS*Only12b1                                4.76E-06          0.36    9.46E-06            0.81       2.22E-06      0.17
FLOWS*Other                                   9.15E-06          0.59    1.46E-05            1.01       2.41E-06      0.29
Captive                                      -6.15E-04         -1.44   -8.66E-04           -1.83      -4.56E-04     -1.26
Unaffiliated                                  4.67E-04          1.58   -1.62E-05           -0.03      -3.40E-04     -1.24
Noload                                        3.99E-04          0.83   -2.57E-04           -0.43       1.29E-04      0.49
Only12b1                                     -6.26E-04         -1.03   -8.10E-04           -1.08      -3.47E-04     -0.47
Other                                        -2.55E-04         -0.62   -4.54E-04           -1.60       2.24E-04      0.47
Panel B: Tests
FLOWS*Captive - FLOWS*Unafil=0                2.67E-05          1.90    1.95E-05            2.34       1.07E-06           0.09
FLOWS*Captive - FLOWS*Noload=0                1.07E-05          0.89   -4.44E-07           -0.05      -3.24E-06          -0.28
FLOWS*Captive - FLOWS*Only12b1=0             -1.11E-06         -0.06    4.86E-06            0.34       1.37E-05           0.79
FLOWS*Unafil-FLOWS*Noload=0                  -1.60E-05         -2.85   -2.00E-05           -3.88      -4.31E-06          -0.78
FLOWS*Unafil - FLOWS*Only12b1=0              -2.78E-05         -2.22   -1.47E-05           -1.16       1.26E-05           0.97
FLOWS*Noload - FLOWS*Only12b1=0              -1.18E-05         -0.87    5.30E-06            0.51       1.69E-05           1.37
Captive - Unafil = 0                         -1.08E-03         -1.89   -8.49E-04           -1.42      -1.16E-04          -0.20
Captive - Noload=0                           -1.01E-03         -1.41   -6.08E-04           -0.98      -5.85E-04          -1.11
Captive - Only12b1=0                          1.10E-05          0.01   -5.60E-05           -0.06      -1.09E-04          -0.15
Unafil - Noload=0                             6.76E-05          0.17    2.41E-04            0.89      -4.69E-04          -1.26
Unafil - Only12b1=0                           1.09E-03          1.42    7.93E-04            0.82       6.79E-06           0.01
Noload - Only12b1=0                           1.03E-03          1.15    5.52E-04            0.55       4.76E-04           0.54




                                                               32
                            Table 6. Recapture of Flows in the Same Complex

This table provides regressions of the predicted redemptions from a fund in the complex on the surprise inflows into all
other funds in the complex. The sample includes all fund classes and has 205,259 fund/month observations where one fund
observation groups all shareclasses together. Funds are removed if there are less than three funds in the complex. The
residual and predicted percent flows are estimated using the specification in Table 4 except flows are defined over current
TNA rather than lagged TNA. The residual dollar flows into the complex are calculated by multiplying the residuals of the
regression model by TNA and then summing these dollar residuals across the complex (excluding the current fund’s
residual flows). Dollar flows to the complex are then divided by the total net assets of the complex (excluding the current
fund’s TNA) to get the percent. NOLOAD represents funds that are not subject to a load and do not pay a 12b1 fee.
CAPTIVE is defined as a fund paying a front load which is allocated to either the underwriter or captive sales force but
none is allocated to an unaffiliated broker. UNAFFILIATED is one if a front load is paid and is allocated to the unaffiliated
broker but not to the captive sales force. ONLY12b1 is defined as a fund charging no load but there is still an allocation to
the broker from 12b1 fees. OTHER funds include all remaining funds. HIGH is defined as 1 if a fund is above the median
performance for all funds in the same objective category in that month and year. LOW are all those funds falling below the
median performance. To adjust for autocorrelation in the panel, we repeat this regression separately for each month and
year between 1996-2003. We report the estimated intercept from regressing the individual coefficients on a constant and
report the t-statistic of this regression using Newey-West standard errors with 5 lags.

                                       Residual Flows into the Complex (excluding                 current fund)
                                         All                     High                                    Low
                                 Coef         t-stat       Coef       t-stat                      Coef          t-stat
Predicted Pctred Captive            0.664        4.475        0.963      2.612                       0.531         4.522
Predicted Pctred Unafil            -0.129       -2.075       -0.236     -1.190                      -0.267        -6.731
Predicted Pctred Noload            -0.188       -2.460       -0.180     -0.676                      -0.384        -3.578
Predicted Pctred Only12b1           0.120        0.260        0.782      1.810                       0.026         0.058
Predicted Pctred Other              0.588        4.186        1.349      4.972                       0.528         3.072
Log Complex Size                   -1.652       -3.454       -2.743     -3.624                      -0.421        -0.780
Log Fund Size                      -1.166       -5.321       -0.490     -2.217                      -1.739        -5.569
Captive                            14.110        3.312        7.650      0.662                      13.151         2.921
Unafil                             26.444       11.363      35.474      10.990                      26.467         8.505
Noload                             33.063        7.168      37.051       3.716                      38.935         5.848
Only12b1                           19.417        0.582     -15.612      -0.499                      14.065         0.429
Other                              -5.521       -1.000     -27.129      -3.136                     -12.528        -1.608
Panel B: Tests of Differences in Interaction Coefficients
Interaction with Pctred           Diff        t-stat       Diff       t-stat                       Diff           t-stat
 Captive = Unafil                     0.79         4.99        1.20        3.59                       0.80             6.52
 Captive = Noload                     0.85         4.93        1.14        3.36                       0.91             6.63
 Captive = Only12b1                   0.54         1.19        0.18        0.35                       0.50             1.03
 Unafil = Noload                      0.06         0.56       -0.06       -0.31                       0.12             1.02
 Unafil = Only12b1                   -0.25        -0.57       -1.02       -2.26                      -0.29            -0.66
 Noload = Only12b1                   -0.31        -0.64       -0.96       -1.75                      -0.41            -0.88




                                                              33
                          Table 7. Rebalancing of Flows in the Same Complex

This table provides regressions of the predicted redemptions from an equity fund in the complex on the surprise inflows
into all other non-equity funds in the complex. We define an equity fund as one with an ICDI objective category of AG,
LG, or GI. Non-equity funds are the remaining funds. The sample includes all fund classes and has 205,259 fund/month
observations where one fund observation groups all shareclasses together. Funds are removed if there are less than 3 funds
in a complex. The residual and predicted percent flows are estimated using the specification in Table 4 except flows are
defined over current TNA rather than lagged TNA. The residual flows into the non-equity funds of the complex are
calculated by multiplying the residuals by TNA and then summing these dollar residuals across all non-equity funds in the
complex. Dollar flows to the complex are then divided by the total net assets of all non-equity funds in the complex to get
the percent. NOLOAD represents funds that are not subject to a load and do not pay a 12b1 fee. CAPTIVE is defined as a
fund paying a front load which is allocated to either the underwriter or captive sales force but none is allocated to an
unaffiliated broker. UNAFFILIATED is one if a front load is paid and is allocated to the unaffiliated broker but not to the
captive sales force. ONLY12b1 is defined as a fund charging no load but there is still an allocation to the broker from 12b1
fees. OTHER funds include all remaining funds. HIGH is defined as 1 if a fund is above the median performance for all
funds in the same objective category in that month and year. LOW are all those funds falling below the median
performance. To adjust for autocorrelation in the panel, we repeat this regression separately for each month and year
between 1996-2003. We report the estimated intercept from regressing the individual coefficients on a constant and report
the t-statistic of this regression using Newey-West standard errors with 5 lags.

                                        Residual flows into Non-Equity funds in the complex
                                 All equity funds         High equity funds       Low equity funds
                                Coef         t-stat       Coef        t-stat      Coef        t-stat
Predicted Pctred Captive           1.438        3.300       2.058        2.239       0.985       3.046
Predicted Pctred Unafil            0.048        0.345      -0.810       -2.158      -0.010      -0.055
Predicted Pctred Noload           -0.431       -2.452      -1.396       -2.748      -0.527      -1.729
Predicted Pctred Only12b1          1.165        3.442       0.870        1.341       1.043       2.550
Predicted Pctred Other             0.175        0.728       0.586        1.067      -0.172      -0.450
Log Complex Size                   1.724        1.639       4.341        2.905       2.245       2.049
Log Fund Size                     -4.477       -9.040      -5.514       -9.277      -4.419      -6.487
Captive                          -11.307       -0.800     -51.833       -1.659       2.148       0.196
Unafil                            21.199        3.060      34.822        3.814      20.549       2.201
Noload                            42.247        5.215      73.197        4.112      46.242       3.206
Only12b1                         -55.881       -2.392     -49.383       -1.337     -55.122      -1.830
Other                              9.208        0.860     -23.547       -1.382      20.457       1.228
Panel B: Tests of Differences in Interaction Coefficients
Interaction with Pctred          Diff        t-stat       Diff        t-stat       Diff       t-stat
 Captive = Unafil                    1.39         3.21        2.87         2.91       0.99         2.62
 Captive = Noload                    1.87         3.57        3.45         3.46       1.51         3.21
 Captive = Only12b1                  0.27         0.53        1.19         1.35      -0.06        -0.12
 Unafil = Noload                     0.48         3.33        0.59         2.55       0.52         2.03
 Unafil = Only12b1                  -1.12        -3.36       -1.68        -2.82      -1.05        -2.59
 Noload = Only12b1                  -1.60        -4.28       -2.27        -3.36      -1.57        -3.04




                                                             34
                                                        Table 8. Cannibalizing of flows within the same complex
This table provides two regression models. The first regresses a fund’s performance ranking for each month/year on the surprise inflows into all other funds in the complex. The second model
regresses the surprise inflows into a fund on the surprise inflows into all other funds in the complex. The sample includes all fund classes and has 205,259 fund/month observations where one fund
observation groups all shareclasses together. Funds are removed if there are less than 3 funds in a complex. The residual and predicted percent flows are estimated using the specification in Table 4
except flows are defined as a percent of current TNA rather than lagged TNA. Surprise inflows into the all funds of the complex are calculated by multiplying the residuals of the regression model
by TNA and then summing these dollar residuals across all funds in the complex excluding the current fund. We then divide these dollar residuals by the total net assets of all the funds in the
complex excluding the current fund to get the residuals as a percent of total net assets. NOLOAD represents funds that are not subject to a load and do not pay a 12b1 fee. CAPTIVE is defined as a
fund paying a front load which is allocated to either the underwriter or captive sales force but none is allocated to an unaffiliated broker. UNAFFILIATED is one if a front load is paid and is
allocated to the unaffiliated broker but not to the captive sales force. ONLY12b1 is defined as a fund charging no load but there is still an allocation to the broker from 12b1 fees. OTHER funds
include all remaining funds. HIGH is defined as 1 if a fund is above the median performance for all funds in the same objective category in that month and year. LOW are all those funds falling
below the median performance. To adjust for autocorrelation in the panel, we repeat this regression separately for each month and year between 1996-2003. We report the estimated intercept from
regressing the individual coefficients on a constant and report the t-statistic of this regression using Newey-West standard errors with 5 lags.

                                                                 Residual Flows into the Complex (excluding current fund)
                                                     All                                                      All                                   High                         Low
                                          Coef         t-stat                                          Coef         t-stat                     Coef      t-stat            Coef      t-stat
          Rkret*Captive                   -15.885        -3.650 Residual Pctnew*Captive                 -0.054        -3.110                    -0.078     -3.702           -0.036     -1.393
          Rkret*Unafil                      -3.429       -1.374 Residual Pctnew*Unafil                   0.048         4.087                     0.051      4.819            0.045      2.889
          Rkret*Noload                      -3.865       -1.219 Residual Pctnew*Noload                   0.199        23.657                     0.189     16.908            0.221     14.108
          Rkret*Only12b1                   27.630         3.587 Resdiual Pctnew*Only12b1                 0.278         8.143                     0.286      9.837            0.304      4.363
          Rkret*Other                       -3.835       -1.532 Residual Pctnew*Other                    0.030         3.929                     0.041      4.071            0.016      1.556
          Log Complex Size                  -1.211       -2.912 Log Complex Size                        -0.735        -1.959                    -1.271     -4.259           -0.092     -0.184
          Log Fund Size                     -1.305       -6.261 Log Fund Size                           -1.789        -9.331                    -1.531     -9.261           -1.933     -6.159
          Captive                          46.935         8.896 Captive                                 37.166         9.819                    37.686      9.893           34.824      6.921
          Unafil                           19.305         5.627 Unafil                                  15.874         6.728                    18.485      6.378           11.616      3.945
          Noload                           21.600         7.170 Noload                                  18.076         8.389                    19.827      6.644           15.446      5.317
          Only12b1                         11.115         2.886 Only12b1                                19.530         8.557                    24.702      8.072           13.543      3.653
          Other                            19.241         8.508 Other                                   15.777         9.171                    18.147      6.873           11.525      4.841
          Panel B: Tests of Differences in Interaction Coefficients
          Interaction with Rkret          Diff         t-stat   Interaction with Residual Pctnew       Diff         t-stat                     Diff          t-stat         Diff         t-stat
           Captive = Unafil                 -12.46         -2.41 Captive = Unafil                         -0.10         -4.46                    -0.13           -5.48        -0.08          -2.23
           Captive = Noload                 -12.02         -1.82 Captive = Noload                         -0.25       -15.11                     -0.27         -11.40         -0.26          -7.71
           Captive = Only12b1               -43.52         -6.21 Captive = Only12b1                       -0.33         -7.90                    -0.36           -9.75        -0.34          -4.35
           Unafil = Noload                    0.44          0.17 Unafil = Noload                          -0.15         -9.63                    -0.14           -7.45        -0.18          -9.19
           Unafil = Only12b1                -31.06         -4.14 Unafil = Only12b1                        -0.23         -7.25                    -0.23           -9.03        -0.26          -3.58
           Noload = Only12b1                -31.49         -3.51 Noload = Only12b1                        -0.08         -2.16                    -0.10           -3.25        -0.08          -1.04




                                                                                                  35
                 Table 9. Flows as a function of payment to captive and unaffiliated brokers

This table provides regressions of percent of TNA in redemptions, inflows, and net flows on ranked returns of the fund.
These only include equity funds classified as AG, LG, or GI in CRSP and funds with multiple shareclasses are grouped
together. REDEMPTIONS are the dollar value of redemptions leaving a funds in a given month divided by the total net
assets of the fund 12 months earlier. INFLOWS are the total dollar value of new money coming into a fund in a given
month divided by the total net assets of the fund 12 months earlier. NET FLOWS are the dollar inflows plus reinvestments
less redemptions in a given month for a fund divided by its total net assets 12 months earlier. We define a fund as having a
CAPTIVE sales force if there is a positive amount of front load allocated to paying a captive sales force while none of the
front load is allocated to an unaffiliated broker. We define a fund as distributing through an UNAFFILIATED broker if
there is a positive amount of front load allocated to an unaffiliated broker while none of the front load is allocated to a
captive sales force. RKRET is the ranked return of a fund over the past 12 months where the ranking is done for every
month and compares all the funds in the sample for that month ranking them from 0 to 1. HICAP is one of the amount paid
to the captive sales force is above the mean paid to captive brokers in that month and zero otherwise. HIUNAF is one if the
amount paid to an unaffiliated broker is above the mean paid to these agents for that month and zero otherwise. Panel A
and C examine the marginal effect of above median compensation to captive brokers. Panel B and D examine the marginal
effect of above median compensation to unaffiliated brokers. Also, Panels A and B examine the results for the low
performers or the bottom 50% of ranked returns. Panels C and D examine the results for the high performers, or the top
50% of ranked returns. To adjust for autocorrelation in the panel, we repeat this regression separately for each month and
year between 1996-2003. We report the estimated intercept from regressing the individual coefficients on a constant and
report the t-statistic of this regression using Newey-West standard errors with 5 lags.


    Panel A: Level of payment to captive brokers (Low Performance)
    DepVar                     Redemptions                    Purchases                           Netflows
                           Coef            t-stat        Coef          t-stat                Coef          t-stat
    RkRet                     -16.72            -2.28        21.15           2.47               46.09            5.52
    RkRet*HICAP                -4.79            -0.42        14.65           0.94               26.10            1.93
    HICAP                      -9.39            -3.35       -10.25          -2.08               -2.47           -0.58
    Intercept                  37.85           10.66         24.04           6.49              -11.42           -3.36
    Panel B: Level of payment to unaffiliated brokers (Low Performance)
    DepVar                     Redemptions                    Purchases                            Netflows
                           Coef            t-stat        Coef          t-stat                Coef             t-stat
    RkRet                     -35.80            -2.80         1.19           0.08               39.27               5.52
    RkRet*HIUNAF               23.66             1.51        49.04           1.97               28.59               1.87
    HIUNAF                    -18.34            -3.81       -15.97          -2.56                1.86               0.70
    Intercept                  54.17             9.18        42.33           6.45               -8.22              -2.80
    Panel C: Level of payment to captive brokers (High Performance)
    DepVar                     Redemptions                    Purchases                           Netflows
                           Coef            t-stat        Coef          t-stat                Coef          t-stat
    RkRet                      46.45             4.72      140.10            6.12               99.39            4.93
    RkRet*HICAP               -15.58            -0.93       -30.39          -0.91                1.47            0.06
    HICAP                      -5.78            -0.56         5.06           0.24               -1.37           -0.08
    Intercept                   5.82             0.91       -40.64          -2.92              -44.55           -3.60
    Panel D: Level of payment to unaffiliated brokers (High Performance)
    DepVar                     Redemptions                    Purchases                           Netflows
                           Coef            t-stat        Coef          t-stat                Coef          t-stat
    RkRet                      30.45             2.78        82.67           6.76               63.67            7.47
    RkRet*HIUNAF                5.97             0.60        53.78           2.06               44.78            2.03
    HIUNAF                    -13.05            -1.94       -27.77          -1.51              -12.69           -0.80
    Intercept                  22.37             3.11         3.18           0.35              -21.06           -3.15


                                                             36

				
DOCUMENT INFO