LIBRARY
OF THE
MASSACHUSETTS INSTITUTE
OF TECHNOLOGY
ALFRED P. SLOAN SCHOOL OF MANAGEMENT
FACTORS THAT AFFECT MUTUAL FUND GROWTH
256-67
F. B. Allderdice and D. E. Farrar
MASSACHUSETTS
INSTITUTE OF TECHNOLOGY
50 MEMORIAL DRIVE
CAMBRIDGE, MASSACHUSETTS 02139
FACTORS THAT AFFECT MUTUAL FUND GROWTH
F. B. Allderdice and D. E. Farrar
Sloan Fellow and Associate Professor of Finance, respectively, at the
Sloan School of Management, Massachusetts Institute of Technology.
WAY 01 ^'-^^^
M. I. T. UbWAKicb_
FACTORS THAT AFFECT MUTUAL FUND GROWTH
Introduction
The substantial growth of the mutual fund industry during the
last few years has attracted the attention of students of finance, eco-
nomics and public policy alike. Net assets managed by such funds have
grown from approximately $^50 million in 19'+0 to more than $38 billion
by June of 1966. During 1965 the mutual fund industry funneled some
$5.2 billion of new (primarily equity) funds into the capital markets;
more than twice the $2.3 billion in new equity raised by all non- financial
U. S. corporations during the year. Growth of the industry has not
been uniform, however, but has been concentrated among a relatively
small number of highly successful funds.
A portion of the industry's growth can be traced to apprecia-
tion in the market value of the securities under its management. Another
and far larger portion, however, represents net new investment by the
public in mutual fund shares. Reasons for the latter source of growth,
and especially for its uneven distribution among the industry's parti-
cipants, are examined in the present study.
Methodology
Broadly speaking, any of a wide variety of forces may be ex-
pected to have an impact on mutual fund growth. Some of these, such
Securities and Exchange Commission, Public Policy Implication of
Investment Company Growth, 89th Congress^ Second Session, House Report
#2337, U. S. Government Printing Office, Washington, D. C, I966.
â– x-x-
Ibid .
Ibid.
as growth in gross national product, stock prices, and disposable
personal income may be expected to be largely time- related phenomena,
and to bear relatively uniformly on all members of the industry. Others,
such as fund performance, sales effort, and size may be expected to
vary widely across funds in the industry. Cross-section analyses,
rather than the analysis of economic aggregates over time, are required
to detect the latter source of variation in mutual fund growth. By
using successive cross sections over time, however, an effort also is
made to control at least roughly for important, time-dependent phenomena.
Simple correlations between net, new money inflows and any of
a broad range of potential explanatory variables can be helpful in iso-
lating those factors that, in a statistical sense, appear most closely
related to mutual fund growth. Care must be taken, however, to avoid
confusing statistical association with causality. There is, clearly,
no fool-proof guarantee that such confusion may be avoided. By simul-
taneously examining the relationship between a dependent variable (such
as net new investment in mutual fund shares) and as broad a range as
possible of potential explanatory variables (such as performance, sales
effort and general economic growth), the danger of spurious association
can, at least, be reduced. Accordingly, a least squares regression
model that simultaneously measures partial relationships between depen-
dent and independent variables is employed in the present study.
Defining:
_Y as a vector of observations on net new money, the
variable whose behavior is to be explained.
X as a matrix of observations on a set of explanatory
variables thought to affect the flow of new funds
into an investment company,
_b as a vector of structural parameters that relates
expected values of _Y to X, and
U as a vector of stochastic error terms,
the basic relationship hypothesized can be summarized as
Y = X Jb + U.
Measures of estimation efficiency can be summarized through
numerous statistics. A common measure of an equation's explanatory
2
power, R , may be interpreted as the percentage of a dependent variable's
total variation that can be explained through a given set of explanatory
variables. Similarly, each independent variable's contribution to an
equation's explanatory power often is measured by the familiar t- ratio,
t = b/a^,
where b is the variable's regression coefficient, and ct, is the coeffi-
2
cient's sample standard deviation. Measures of R greater than .10
or .20 are sufficiently large to suggest the presence of significant
(non-zero) relationships between Y and X over cross sectional samples
such as those encountered here; while t-ratios greater than 1.0 or 2.0
generally are interpreted as identifying meaningful (non-zero) partial
relationships between dependent and independent variables. Squared
multiple correlation coefficients, regression coefficients, and t-
ratios, accordingly, are reported in the analyses that follow.
Scope of the Study
This study's — indeed any study's - scope can be defined
along at least three dimensions: its units of observation (in this
case mutual funds), the time period over which observations are col-
lected, and the variables by which each observation is charaterized.
The present study employs observations on I69 mutual funds,
approximately U5^ of the funds in existence during June, I966. A
better idea of the sample's significance in terms of the industry's
overall size, however, can be gained by noting that the companies
covered received some $3.2 billion in net new investment during 1965?
more than 60'^^ of the industry's new money during the year. In terms
of total assets managed, this sample contained $30-2 billion at the
end of 1965, or 79^ of the industry's total assets. Similarly, com-
panies covered by the present sample contain 5*9 million shareholder
accounts, or 88^ of the industry's accounts. By any measure, therefore,
the sample covers a significant portion of the investment company
industry.
Data is collected on each fund over a three-year period of
time, from 1963-65. Data limitations, especially regarding portfolio
turnover, constitute the primary constraint on the sample's coverage.
Data elements by which each fund and year are characterized
include percentage increases in gross national product, total personal
income, and Standard and Poor's 5OO Stock Security Price Index, plus
9 measures specific to each fund, summarized in Table I. Most of the
variables in the Table are self-explanatory. Total net assets and
numbers of shareholders constitute rough indications of a fund's size.
- 5 -
Table I
RAW DATA
Factor Description
Symbol
Total Wet Assets
TA
Net Asset Value
NAV
Portfolio Turnover Ratio
pro
Dividends
D
Capital Gains
CG
Shareholders
s
Expense Ratio
ER
Sales Charge
SO
Fund Objective
OBJ
- 6 -
Net asset value, dividends and capital gains distributions provide
the per-share data out of which various performance measures are formed.
Expense ratio is defined as the ratio of expenses actually paid by the
fund, as a per cent of total net assets. It includes management fees,
the salaries of fund employees, office expenses and other expenses not
provided by the fund's advisor. Expense ratio, however, does not in-
clude services provided to the fund or its advisor in return for broker-
age. Sales charge is defined as the maximian load, applicable to a
fund's smallest (non- contractual plan) investors. Portfolio turnover
is calculated by dividing the lesser of sales or purchases, excluding
government securities and other short-term obligations, by the average
of total net assets over a year; and. is the same as that reported to the
Securities and Exchange Commission on Form N-IR. Fund objective is
that adopted by Arthur Weisenburger and Company.
From the raw data described above, a series of working data
elements is developed for each observation, and summarized in Table II.
Certain variables - specifically, total gain, dividend yield, capital
gains distribution, unrealized gain, expense ratio and portfolio turn-
over - are lagged by one year to more closely approximate the informa-
tion available to an investor at the time of a fund purchase decision.
A complete set of raw and working data elements is summarized for a
single fund in Tfeble III.
Securities and Exchange Commission, Form N-IR, Annual Report of Registered
Management Investment Companies under the Investment Company Act of 19^0
and the Securities and Exchange Act of 193^, January 25, 1965; item 1.25(a).
Wessman, L., Editor, Investment Companies , Arthur Weisenburger and
Company, New York, I966.
- 7 -
Table II
Working Data
Variable Symbol
Net New Investment, as per cent of
total assets NNI^
Net New Investment, per cent, lagged
one year NNI^, ,
Percentage growth in Standard
and Poor 500 Stock Index SP
Dividend Yield, lagged one year ^\-l
Capital Gain Yield, lagged one year "^"^^-l
Unrealized Gain, lagged one year ^*^t-l
Total Gain, lagged one year '^t-1
Portfolio Turnover, lagged one year PTO, _,
Sales Charge SC
Expense Ratio, lagged one year ^^t-l
Total Assets, lagged one year
(in logarithms) "^t-l
Fund Objective OBJ
o
o o o o
o o o o
I OJ ON rH
^Jb • • •
M CVJ VD t^
ITN UA UA Lf>
■• • •
CO CO CO CO
CVJ CO o o
CO f- CO CO
M3
CO
H
LTN
CO
CO
H
VD
CQ
m
r-i
H
o>
1^
t~-
t-
ir\
H
H
H
iH
t:I
-:t
O
u^
u^
a
C5
C\J
oo
en
VD
M
M
^
n3
O
o
o
o
•
O
M
a
S
0)
0)
H
VD
t^
CO
en
,Q
Td
&
>H
m
oo
oo
-d-
r"'
Jh
O
o
O
o
o
Eh
<
•
ft
«
CO on po oo
LTN oj m CO
^
J-
H
rH
r-i
g
-4-
H
1
LTN
•
o
O
H
CVJ
d
>
O
ON
00
ON
ON
OJ OJ OJ CXJ
^
ro
^
-^
VD
O
OJ
t-
UA
o
o
CO
CVJ
CO
Lr\
ro
OJ
OJ
oo
J-
CVJ
ro
J-
l/N
vo
VO
vo
VO
CJN
C7N
ON
ON
H
rH
H
H
a
O
H
o
3
o
O
O
o
rH
H
O
f-
1
-4-
O
oo
<
OJ
J-
VO
P
o
OJ
-d-
0)
•
•
•
&D
o
o
O
O
r-\
rA
r-\
h:i
rH
CO
OO
oo
LTN
CVJ
oo
P
^
OO
VD
S
rH
rH
-1
On
-^
-^
o
o
r^
rH
r-t
LTN
r-K
CO
1
^
OO
l/A
>H
a
«
Q
rH
H
H
CVJ
r-\
-^
rH
C\J
-^
O
1
m
J-
o
g
LTN
o
rH
1
CVJ
rH
N-
CVJ
ON
â– &'5-
-cf
CVJ
J-
H
CJN
r-{
ON
^z^
g
VD
I^
OO
-^
ITN
VO
VO
VD
On
rH
ON
rH
On
- 9 -
Some of the variables described above may require further
comment. Although it would be desirable to characterize each fund and
year by a complete spectriom of the variables that affect net new money
inflows during the year, it must be recognized that any characteriza-
tion will, in fact, be incomplete. To reduce the danger of introducing
spurious relationships between measured variables and other character-
istics that should be, but are not included in an independent variable
set, an effort is made to proxy excluded characteristics (such as the
size and efficiency of a fund's sales force), by including in each
equation a lagged value of its dependent variable. The rationale
underlying such a variable's use may be traced to a belief that factors
excluded from a given year's set of independent variables are likely
to have been operative during the preceding year, as well. Analyses
of the relationships reported below, excluding this effect, indicate
the presence of serial correlation between observed residuals, and
*
accordingly, support such a variable's use.
A second broad class of measurement problems concerns the
quantification of variables, such as a fund's objective, that do not
lend themselves to continuous measurement. A fund either is a growth
fund or is not; or a maximum performance fund, or not; or an income
fund, or not, etc. To the ext.ent that labels possess sales appeal, an
ability to measure the appeal inherent in a fund's objective may be
desirable. Accordingly, a set of dummy variables is defined for each
of k Weinsenburger objectives: maximuiri gain, growth, growth- income.
For further discussion of this problem, see L. M. Koyck, Distributed Lags
and Investment Analysis , North Holland Publishing Company, Amsterdam, 195^;
and L. R. Klein"^ "The Estimation of Distributed Lags," Econometrica , 26,
^, 1958«
- 10 -
and income. Should a fund fall into one of these categories the vari-
able corresponding to the category assumes a value of 1.0, if not its
value is 0. Each fund, of course, falls into one and only one category.
To avoid singularity, of course, one objective (in this case the Growth-
Income- Stability amalgam; must be excluded from the set. Coefficient
estimates for included objectives, accordingly, are interpreted as
the difference between a given objective's sales appeal and that enjoyed
by a Growth- Income- Stability investment objective.
An effort also is made to reduce the enormous variation in
new money attributable only to fund size., by transforming total asset
measures to logarithms, and standardizing each equation's dependent
variable, net new investment, to a percentage of the fund's beginning-
of-year total assets.
Analyses
As indicated earlier, the study is designed to investigate
the relationship between percentage growth in investment company assets
through net new investment, and a wide variety of both general economic
conditions and specific fund characteristics. Broadly speaking, five
general categories of factors are investigated: economic conditions,
sales effort, fund performance, fund size, and fund objective. Initial
exploratory analyses eliminate percentage growth in such general economic
characteristics as gross national product and total personal income
See D. Suits, "Use of Dummy Variables in Regression Equations, " Journal
of the American Statistical Association, 52, 28, 1957.
11
from further consideration. An example of the type of finding uncovered
is summarized in Table IV, where all k68 usable observations on l69 mutual
funds are pooled in a single regression model. Four separate equations,
each employing a different measure of performance (total gain, dividend
yield, capital gains yield, and unrealized gain) are summarized. Regression
2
equations, t-ratios, R corrected for degrees of freedom, and sample size
are reported for each equation.
Each equation's statistical significance is considerable. Differ-
2
ences in R are trivial, leading to no preference for one equation, and
therefore one measure of performance, over another. Comparisons between
t-ratios for different parameters, however, highlight the overwhelming
importance of lagged net new investment as an explanatory variable; sug-
gesting the existence of both relevant, excluded, independent variables
and considerable stability in the forces that determine the distribution over
time of new money inflows between companies in the industry. The lagged
variable ' s considerable explanatory power also tends to reduce confidence
in the validity of equations (over this set of data) that do not contain
such a variable.
Portfolio turnover, fund size, expense ratio, and fund performance
also appear to be highly signficant determinants of the distribution of
net new investment between competing funds. Certain fund objectives such
as growth and grovrth- income also appear to have at least moderate investor
appeal. Market action, measured by percentage change in the Standard and
Thirty- nine missing data elements, primarily portfolio turnover, account
for the discreptancy between k68 useable observations and 3 x l69 = 507
expected data points for 169 funds over a three-year period of time.
s^
OJ
CVJ
ON
CTn
CM
cu
SI
12
|c
to
J- 0^
OJ OJ r-i r-i
o o o o
21^
H 0|
1 H -^
^
p
P<
u
(1
Qi
m
i;
Tl
o
fl
•H
5
â– rt
ti
a
(11
h-l
•H
p.
-rt
(1)
x:
:3
• +
' +
MD
'
.+ .1
+ •+ •+ +
.+
CJ
3
.+
,
+
\ /
•+ • +
•+ 1 + •+
• +
H
03
O
1
, ,
+
V
•+ • +
•+ 1 + •+
• +
O
-=t
m
g
Q
+
.1
• +
A
•+ • +
• + 1 + +
• +
t^
OJ
E-i
. +
+
•+ /
/ \
r+ • +
• + 1 + • +
• +
>>
V ,
+> â– o
3
• +
•+
1
+
•+ 1
\ /•+ •+
• +
H -p
â– H C
a] n-l
o