The Persistence of Mutual Fund Performance

Author(s): Mark Grinblatt and Sheridan Titman

Source: The Journal of Finance, Vol. 47, No. 5, (Dec., 1992), pp. 1977-1984
Published by: Blackwell Publishing for the American Finance Association
Stable URL: http://www.jstor.org/stable/2329005
Accessed: 20/05/2008 08:34

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=black.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We enable the
scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that
promotes the discovery and use of these resources. For more information about JSTOR, please contact [email protected].

http://www.jstor.org
THE JOURNAL OF FINANCE * VOL. XLVII, NO. 5 . DECEMBER 1992

The Persistence of Mutual Fund

Performance

MARK GRINBLATT and SHERIDAN TITMAN*

ABSTRACT
This paper analyzes how mutual fund performance relates to past performance.
These tests are based on a multiple portfolio benchmark that was formed on the
basis of securities characteristics. We find evidence that differences in performance
between funds persist over time and that this persistence is consistent with the
ability of fund managers to earn abnormal returns.

THERE IS A LARGEand growing industry devoted to measuring mutual fund

performance. This industry, as evidenced by newsletters, regular features in
the financial press, and the existence of firms that professionally evaluate
funds, is based on the idea that funds that do well (or poorly) in the past will
continue to do so in the future.
While the efficient markets hypothesis has generated interest among aca-
demics in testing whether funds exhibit performance, there has been very
little research devoted to testing for persistence in fund performance.1 One
reason for this may be that the traditional benchmarks used to evaluate
performance are known to exhibit persistent biases. For example, the CAPM
and APT-based benchmarks favor small capitalization and high dividend-yield
stocks. Thus, small-firm funds and income-oriented funds may appear to
persistently.outperform other funds when traditional benchmarks are used.
In a recent paper,2 we introduce a new benchmark that does not exhibit
these biases and find evidence of significant differences in performance
between funds. Our tests3 reject the hypothesis that all funds with growth
and aggressive-growth investment objectives have equal performance. This
rejection holds for both the actual returns of the funds and for portfolio
returns constructed from the funds' quarterly holdings that proxy for the
funds' gross returns (before transaction costs and fees).

*Anderson Graduate School of Management, UCLA. Titman is also from The Hong Kong
University of Science and Technology. We wish to thank James Brandon and Erik Sirri for
research assistance on earlier versions of this research and the editors and referees, as well as
Russ Wermers, for their comments.
1 Exceptions include Jensen (1969), Beebower and Bergstrom (1977), and Lehmann and
Modest (1987). All of these studies are subject to a benchmark bias and the latter has no test
statistics.
2 Grinblatt and Titman (1989a).
3 See Tables 2 and 3 of Grinblatt and Titman (1989a).

1977
1978 The Journal of Finance

The F-test used in this earlier study, which is based on a nonspecific

alternative hyopthesis, does not tell us if past performance is a useful guide
for selecting a fund. If we are interested in learning whether past perfor-
mance provides useful information to an investor, more specific tests of the
relation between past performance and future performance are needed. Such
tests are provided in this paper. In contrast to the earlier paper, we use a
larger sample of funds and focus only on the actual returns of the funds.
The organization of the paper is as follows. Section I describes our sample
of funds, the sample of passive portfolios we use as a control, and the
benchmark we use to evaluate performance. Section II describes our persis-
tence tests and presents the empirical results, and Section III concludes the
paper.

I. The Sample of Fund Returns, the Control Sample, and the

Benchmark
Mutual fund data, consisting of monthly cash-distribution-adjusted returns
and investment goals for 279 funds that existed from December 31, 1974 to
December 31, 1984, were purchased from CDA Investment Technologies, Inc.
To our best knowledge, this is the largest sample of mutual funds evaluated
in the academic literature. The data was spot checked with data collected by
hand and found to be accurate.
As with most mutual fund studies, the mutual fund return data are subject
to survivorship bias. Since CDA's nonacademic clients are only interested in
mutual funds that they can invest in, funds that went out of business prior to
December 31, 1984 are excluded from the CDA data set. Grinblatt and
Titman (1989a) conclude that the survivorship bias is negligible for a sample
that includes surviving and nonsurviving funds over this time period. More-
over, the survivorship requirement biases our tests against finding persis-
tence.4
To check the reliability of our persistence tests, a control sample of 109
passive portfolios is constructed from the CRSP daily returns and daily
master files. Each portfolio equally weights a subset of the CRSP securities
and is rebalanced monthly. Seventy-two of the portfolios, each containing

4 When the sample is split into two halves, we can characterize a fund's performanceinto four
categories, (1) good in the first half, good in the second half, (2) good in the first half, bad in the
second half, (3) bad in the first half, good in the second half, and (4) bad in the first half and bad
in the second half. Cases (1) and (4) are cases of positive persistence, while cases (2) and (3) are
indicative of negative persistence. If these cases are equally likely, one will find no persistence.
Survivorship requirements, however, are most likely to eliminate funds in category (4). This
would bias the remaining funds towards negative persistence. An alternative view, expressed by
Brown, Boetzmann, Jbbotson,and Ross (1991), is that funds that do well in the first half are
likely to be those that choose the riskiest strategies. This leads to a positive bias in measured
persistence since the funds that do well in the past may be less likely to survive in the future
since their riskier strategies make it more likely that they will do extremely poorly. A recent
paper by Hendricks, Patel, and Zeckhauser (1991) examines persistence strategies on a sample
without survival bias and concludes that the bias is small.
 Persistence of Mutual Fund Performance 1979

about 1/12 of the CRSP securities, are formed on the basis of securities
characteristics. These characteristics include firm size, dividend yield, co-
skewness with the monthly rebalanced equally weighted index, interest rate
sensitivity, past returns over the previous three years, and beta computed
against the equally weighted index. For each of the six characteristics, every
CRSP security is grouped into one of twelve passive portfolios based on its
ranking against other CRSP securities with respect to that characteristic. An
additional 37 passive portfolios are formed on the basis of SIC industry
groupings. Each of these is an equally weighted monthly rebalanced portfolio
containing CRSP securities with the same "two-digit" SIC code as of the end
of 1974. Only industries with at least 20 CRSP-listed firms are included.
The performance measure used in this study, an extension of the measure
employed by Jensen (1968, 1969), is computed relative to the eight-portfolio
benchmark, P8, used in Grinblatt and Titman (1989a). The basic idea under-
lying the formation of this benchmark is that various firm characteristics are
correlated with their stocks' factor loadings. As a result, portfolios formed
from stocks grouped by securities characteristics can be used as proxies for
the factors. The P8 benchmark, constructed from groupings of the passive
portfolios' returns described above, consists of four size-based portfolios, three
dividend-yield-based portfolios, and the lowest past returns portfolio: The
smallest 81/3% of firms comprise the first size-based portfolio; the average of
the second- and third-smallest size portfolios (out of 12) comprise the second
portfolio; the average of the fourth- through ninth-smallest size portfolios
comprise the third portfolio; and the average of the three largest size portfo-
lios comprise the fourth. The three portfolios formed from dividend-yield
rankings are an equal weighting of the two lowest dividend-yield portfolios
(out of 12), the fifth- and sixth-lowest dividend-yield portfolios, and the tenth
and eleventh dividend-yield portfolios. To compute an abnormal return rela-
tive to the P8 benchmark, we use ordinary least squares to estimate the
intercept in a time-series regression of the excess returns (above a one-month
T-bill rate) of the fund or passive portfolio on the excess returns of the eight
portfolios described above.5

II. Persistence Tests

The persistence of abnormal performance is analyzed with a three-step

procedure. First, we split the ten-year sample of fund returns into two
five-year subperiods. Second, we compute the abnormal returns of each fund
for each five-year subperiod. Finally, we estimate the slope coefficient in a
cross-sectional regression of abnormal returns computed from the last five
years of data on abnormal returns computed from the first five years of data.
A significant positive t-statistic for the slope coefficient in this regression
would reject the null hypothesis that past performance is unrelated to future

5For a theoretical discussion of the Jensen Measure and the inferences that can be drawn
from it, see Grinblatt and Titman (1989b).
1980 The Journal of Finance

performance and support the alternative hypothesis that past performance is

positively related to future performance. However, the standard method for
computing the t-statistic for the slope coefficient in a cross-sectional regres-
sion results in a statistic that does not have a true t-distribution. This is
because many funds have similar portfolios and hence have highly correlated
residuals. To overcome this bias, we develop an alternative t-test that is
derived from a time series procedure. The procedure for computing these
"time-series t-statistics" is an extension of the technique introduced by Fama
and MacBeth (1973) to overcome a similar problem in tests of the CAPM.
Define ai to be the abnormal return of the ith fund computed from the first
five years of returns in excess of the average abnormal return of all funds
over these 60 months. By construction, the ai's sum to zero, implying that a
portfolio constructed with weights that are proportional to the ai's has zero
cost. Now consider the following weighted average of the (second-half) re-
turns of the funds,

Rpt =iajRjt/vara where

Rit= the return of the ith fund in month t, t = 61,... , 120, and
var(a)= cross-sectional variance of the abnormal returns of the funds
computed from the first five years times 279.

Since the weight on fund i is ac/var(a), we can regard Rpt as the return (in
the second five years) of a zero cost portfolio of the funds. The intercept from
a multiple regression of Rpt on the excess returns of the eight portfolios in
the P8 benchmark is the abnormal return of this portfolio (or, alternatively,
the same weighted average of the abnormal returns of the individual funds).
This intercept is algebraically identical to the least squares slope coefficient
from the cross-sectional persistence regression. However, in contrast to the
t-statistics estimated with the cross-sectional persistence regressions, the
t-statistics generated with the time-series regression will not be biased under
the null hypothesis that residuals are i.i.d. normal (an assumption that is
better approximated with the time-series regression than with the cross-
sectional regression).6
Panel A of Table I presents the intercept from this time-series regression
along with the corresponding t-statistic. This persistence statistic indicates
that mutual funds in the second five-year period are expected to realize a
0.28% greater abnormal return in the second five years for every 1% abnor-
mal return achieved in the first five years. This coefficient is highly signifi-
cant (at almost the 1% level in a two-tailed test).
As a control, we also performed the same procedure on the sample of 109
passive portfolios described earlier. The coefficient here, also reported in
Panel A, is of a larger magnitude than the coefficient for the 279 mutual
funds, but is statistically insignificant. To reconcile the larger coefficient

6
For most of our results, the time-series t-statistic is about half the size of the biased
t-statistic derived directly from the cross-sectional regression.
 Persistence of Mutual Fund Performance 1981

Table I
Regression Tests of the Persistence of Performance
Panel A: Slope Coefficient and t-Statisticsa for Regressions of Abnormal Returns of 279 Mutual
Funds and 109 Passive Portfolios in the Last 60 Months on Abnormal Returns from the First 60
Months

Sample Cross-Sectional R2 Slope Coefficient t-Statistic

Mutual finds 0.06 0.281 2.64*

Passive portfolios 0.15 0.395 1.59

Panel B: Slope Coefficient and t-Statisticsa for Regressions of the Abnormal Returns of 279
Mutual Funds and 109 Passive Test Portfolios in 60 Randomly Selected Months on the Abnormal
Returns of the Portfolios from the Remaining Sixty Months

Sample Cross-Sectional R2 Slope Coefficient t-Statistic

Mutual funds 0.13 0.420 4.59**

Passive portfolios 0.00 - 0.053 -0.29

Panel C: Abnormal Returns and t-Statisticsa in the Last Sixty Months (or Random Sixty
Months) of Equally Weighted Portfolios of Mutual Funds Consisting of Either the Top or Bottom
10% of the Funds Based on Performance from the Other Half of the Sample

Best Performers Worst Performers t-Statistic

Ranking Period Abnormal Returnsb Abnormal Returnsb for Difference

First 60 months 0.0000 (0.02) -0.0029 (-2.10)* 2.10*

Random 60 months 0.0046 (2.05)* - 0.0028 (-2.31)* 4.32**
a
t-Statistic is unbiased t-statistic calculated with the time-series technique.
b t-Statistics are in parentheses.
* Significant at 0.05 level (two-tailed test).
** Significant at 0.01 level (two-tailed test).

estimated for the sample of passive portfolios with its statistical insignifi-
cance, we have to recognize that these passive portfolios were constructed to
be very different while many of the mutual funds are very similar. As a
result, the returns of the costless portfolio of funds used to calculate the
time-series t-statistics have a much lower standard deviation than the analo-
gous returns constructed from the passive portfolios.
Although the coefficient estimate in the regression for the passive portfolios
in Panel A is not reliably different from zero, its magnitude makes us
hesitant to conclude that the slope coefficient for the funds is entirely due to
persistence of managerial talent. Moreover, recent work by Jegadeesh and
Titman (1991) provides evidence of significant persistence in the long-run
abnormal returns of individual stocks. Panel B of Table I addresses whether
this is the source of persistence for the mutual funds. To obtain the coeffi-
cients in Panel B, we perform the same regressions as in Panel A, but
randomly sort the 120 months of returns into two 60-month samples, sepa-
rately sorting the Januaries so that each subsample would include the same
1982 The Journal of Finance

number of Januaries (this allows us to avoid possible biases due to the

January effect).
The results in Panel B from the randomly partitioned months are even
more striking than those from the chronologically partitioned months in
Panel A. First, there is no evidence of persistence in the passive portfolios.
Moreover, persistence of the mutual funds is much stronger with a random
partitioning of the monthly returns than with a chronological partitioning.
The coefficient estimate, 0.42, is much larger and the t-statistic, 4.59, is
highly significant. This indicates that persistence in individual securities
returns is unlikely to be the source of the persistence observed for either the
mutual funds or the passive funds in Panel A.
The results in Panels A and B for the mutual funds could also have been
generated by persistent differences in fees and transaction costs across funds.
To test this, Panel C examines the average abnormal returns of the 10% best-
and worst-performing funds in a five-year period outside of the ranking
period. The same chronological sorting and random sorting is used. If persis-
tence is entirely due to fees and transaction costs, the best-performing funds
from the ranking period would exhibit negative abnormal returns of approxi-
mately the same magnitude as their transaction costs in the test period.
However, significantly positive abnormal returns for the funds ranked best
would imply that there was persistence in abilities. The t-test in the differ-
ence column of Panel C also tests whether the abnormal returns of the 10%
best-performing funds equals the abnormal performance of the 10% worst-
performing funds in the five-year test period.
Panel C indicates that when the sample is split in half chronologically, the
worst-performing funds demonstrate persistently poor performance; both the
best- and worst-performing funds demonstrate persistence when the fund is
split in half randomly. Funds in the lowest decile from 1975-1979 and in the
random half of the sample used for ranking have abnormal performance of
about - 3.5 percent per year in the other half of the sample.7
In Panel C, the portfolio of funds with abnormal returns in the top decile
for the randomly selected ranking months achieves an abnormal return of
about 5.5 percent per year in the remaining sixty months. However, this
result appears to be due to chance. Out of four additional random partitions
of the 120 months, none resulted in significant out-of-sample abnormal

7
This finding appears to be robust to changes in the random numbers generated by the
computer. Four additional randomly generated sample splits were also examined. Three of the
four had negative returns for the worst-performingfunds of approximatelythe same magnitude
as that for the first random split. These resulted in significant differencesbetween the returns of
the best- and worst-performing funds in the second half of the sample. In addition, the
persistence regressions for the 109 passive portfolios are all insignificant with the P8 bench-
mark. The t-statistic for the difference between the 10% best and worst performers with the
chronological sample split used in Panel C was 1.53. The comparable t-statistics for the five
random splits were respectively: - 0.74, - 1.05, 0.12, 0.23, and - 0.34.
 Persistence of Mutual Fund Performance 1983

returns for the best-performing funds from the ranking period. However, all
of the out-of-sample abnormal returns are positive.8

III. Conclusion

The results presented in this paper indicate that there is positive persist-
ence in mutual fund performance. The persistence cannot be explained by
inefficiencies in the benchmark that are related to firm size, dividend yield,
past returns, skewness, interest rate sensitivity, or CAPM beta. These find-
ings are consistent with there being persistent differences in fees and trans-
action costs across funds, although the results in Panel C suggest that this is
not the sole explanation for our results. Irrespective of the source or sources
of the persistence that we find in this paper, we can assert that the past
performance of a fund provides useful information for investors who are
considering an investment in mutual funds.
One issue that we do not address in this paper is how to optimally weight
information about past performance in selecting a mutual fund. A recent
paper by Hendricks, Patel, and Zeckhauser (1991) argues that only the most
recent past performance provides information about future performance.
They present evidence of higher levels of abnormal performance for strategies
that buy mutual funds based on their performance measured over the past 2
to 8 quarters.

8
The results for the other four random partitions are as follows:

Ranking Best Performers Worst Performers t-Statistic for

Period AbnormalReturns Jensen Measureb Differencea
Random2
60 months 0.0020 (0.91) -0.0031 (2.54)* 2.30*
Random3
60 months 0.0018 (0.88) -0.0037 (-3.21)** 3.26**
Random4
60 months 0.0022 (1.19) -0.0031(-1.97) 2.80**
Random5
60 months 0.0009 (0.48) -0.0003 (-0.17) 0.42

(See Table I for explanation of symbols)

REFERENCES

Beebower, Gilbert L., and Gary L. Bergstrom, 1977, A performance analysis of pension and
profit-sharing portfolios: 1966-1975, Financial Analysts Journal 33, 31-42.
Brown, Stephen, Will Goetzmann, Robert Ibbotson, and Stephen Ross,1991, Survivorship bias in
performance studies, Working paper, Yale University.
Fama, Eugene, and James MacBeth, 1973, Risk, return and equilibrium: Empirical tests,
Journal of Political Economy 71, 607-636.
1984 The Journal of Finance

Grinblatt, Mark, and Sheridan Titman, 1989a, Mutual fund performance:An analysis of quar-
terly portfolioholdings, Journal of Business, 62, 393-416.
, 1989b, Portfolio performance evaluation: Old issues and new insights, Review of
Financial Studies, 2, 393-421.
Hendricks, Darryll, Jayendu Patel, and Richard Zeckhauser, 1991, Hot hands in mutual funds:
Short-run persistence of performance, 1974-88, Working paper, Harvard University.
Jegadeesh, Narasimhan, and Sheridan Titman, 1991, Return to buying winners and selling
losers: Implications for stock market efficiency, Workingpaper, UCLA.
Jensen, Michael, 1968, The Performance of mutual funds in the period 1945-1964, Journal of
Finance, 23, 389-416.
, 1969, Risk, the pricing of capital assets, and the evaluation of investment portfolios,
Journal of Business, 42, 167-247.
Lehmann, Bruce, and David Modest, 1987, Mutual fund performanceevaluations: A comparison
of benchmarks and benchmark comparisons, Journal of Finance, 42, 233-265.