2. International Research Journal of Finance and Economics - Issue 34 (2009) 118
seems that everyone choosing active managers, from pension plan sponsors to individual investors, put
some weight on past performance in portfolio selection.
However, the scholars do not agree on the added-value of the performance evaluation industry
for the community of investors; plenty of studies have found evidence of performance persistence but
there are almost as many studies that have not found it. The topic has fascinated scholars because the
existence of performance persistence would question not only the weak form efficiency of capital
markets, but also that of mutual fund markets. This would imply that abnormal profits over random
fund selection might be earned on the basis of past performance records.
This paper provides the extensive review to the literature of mutual fund performance
persistence from the very first pioneer studies till the most recent advances in tracking performance
persistence. To my knowledge, this is the most comprehensive literature review ever made of the
persistence studies. Besides a review this paper discusses potential explanations for inconsistent
findings of the abundant literature on performance persistence. Also, the time-varying trends of the
persistence literature are presented. This kind of review of the “lessons learned” from the previous
persistence studies may help scholars to improve the research design and hopefully, also the validity of
the forthcoming studies. Also the investment practitioners can exploit the conclusions in their decisions
on mutual fund selection.
Numerous studies that examine performance persistence of institutional portfolios of other
types, such as pension funds, hedge funds, publicly offered commodity funds, and REIT funds have
also been published in the financial journals. For the sake of proliferation of the overall performance
persistence literature, this paper focuses on relative performance persistence of common open-end
mutual funds.
2. The Pioneer Studies of Performance Persistence
The issue of performance persistence is discussed already in the seminal mutual fund studies. The
distinctive feature of the earlier studies is the use of long selection period and typically holding period
of the same length (see Table 1). Sharpe (1966) compares the performance rank orders based on the
Sharpe Ratios of two successive decades and finds positive though not statistically significant
correlation. He also uses the rankings of funds based on the Treynor Ratio computed from the earlier
period data to predict rankings based on the Sharpe Ratio of the later period but the results remain the
same.
Jensen (1968) uses the same lengths of both selection and holding period as Sharpe, but
examines the persistence of abnormal performance determined by the Jensen Alpha. He finds positive
correlation in the performance between the selection period and the holding period indicating that some
funds may be consistently inferior and others consistently superior. However, Jensen emphasizes that
one must be very careful in interpreting these results so that a fund manager who experienced superior
performance in the earlier period would be far more likely to experience superior results in the latter
period. He notes further that positive correlation between these two periods is mainly due to
persistence of inferior performance.
Carlson (1970) analyzes performance persistence of 57 mutual funds on the basis of the sample
data from the time-period of 1948-1967. Splitting this two-decade data like Sharpe and Jensen, he finds
that the interdecade rankings based on the Sharpe Ratio show no evidence of persistence though
rankings based on total return or risk (volatility) do so. As a consequence of his findings that broadly
defined investment objectives might influence performance measurement, Carlson (1970) examines
also smaller sample of 33 funds consisting only of common stock funds but the main results do not
change. Despite the lack of overall consistency in rankings based on the Sharpe Ratio, there appears to
be a slight tendency for funds to remain either in the top or bottom quartiles during both decades.
Carlson further divides each decade into two five-year periods: In general, these five-year rankings
based on the Sharpe Ratio improve dramatically the predictive power of past performance compared to
3. 119 International Research Journal of Finance and Economics - Issue 34 (2009)
ten-year rankings. Also for the observation period of this length, intraquartile statistics shows a strong
tendency for funds to remain within top or bottom groupings.
Sarnat (1972) examines the performance persistence of 56 mutual funds viewing the size of the
efficient sets generated by the four alternative decision criteria including mean-variance criterion.
Using the length of 12 years for both selection period and holding period he concludes that the
composition of the efficient sets over time is not stable enough to benefit an investor economically.
Kritzman (1983) analyzes the performance persistence of 32 Bell System’s fixed-income funds on the
basis of total returns from two successive five-year periods and finds no relationship between past and
future performance even among very best and very worst funds.
Levy and Sarnat (1984) use the same type of the efficient set approach as Sarnat (1972). Using
the data on mutual fund returns for the subperiods of 11 years (1959-69 and 1970-80) the authors
conclude, parallel to the results of Sarnat (1972), that the composition of the efficient sets over time is
not stable enough to derive predictions materially better than simple random choice.
As a part of larger mutual fund performance study Lehmann and Modest (1987) examine the
persistence of fund rankings based on various performance measures (i.e., Treynor & Black (1973)
Appraisal Ratios1, alphas based on both the CAPM model and several applications of the APT models,
and in addition, total returns) for the 15-year period divided further into three 5-year subperiods. The
study of Lehmann and Modest (1987) can be considered one of the cornerstone studies of mutual fund
performance evaluation, since this is the first time when multifactor models are used as the basis of
performance measurement. Although evidence of persistence is found, the authors note that results are
highly dependent on performance metrics employed; the results show considerable differences between
rankings based on the CAPM model and those based on various applications of the APT model.
Moreover, substantial ranking differences occur also within alternative APT implementations.
Lehmann and Modest (1987) stress the need to find a set of benchmarks that represent the common
factors determining fund returns.
1
The Appraisal Ratio is calculated by dividing the Jensen alpha by the nonsystematic risk of that portfolio, i.e., the
standard deviation of the residual term of the regression equation. Also known as the information ratio (e.g., see Grinold
& Kahn, 1995, p. 90), it measures abnormal return per unit of risk that in principle could be diversified away by holding a
market portfolio (Bodie et al., 2005, p. 868).
4. International Research Journal of Finance and Economics - Issue 34 (2009) 120
Table 1: Performance persistence studies of the 1966–1989 period
Table 1 summarizes the main findings of the studies of mutual fund performance persistence
published during the 1966-1989 time period. The common characteristic of the studies of this era is
the use of relatively long selection and holding periods. Noteworthy is also that strongest evidence
of persistence is found systematically in the studies employing shorter selection and/or holding
periods.
Method for Type of funds Time Length of Length of
Authors performance and size of period selection holding Results
evaluation sample period period
Sharpe RatioTreynor
Sharpe 1966 Ratio for SPa/Sharpe 34 mutual funds 1944–1963 10 years 10 years weak evidence of persistence
Ratio for HPb
Jensen 1968 Jensen Alpha 115 mutual funds 1945–1964 10 years 10 years weak evidence of persistence
Treynor Ratio for no persistence based on risk-
57 mutual funds 10 years 10 years
Carlson 1970 SP/Sharpe Ratio for 1948–1967 adj. performance metrics
HP Sharpe Ratio 33 equity funds 5 years 5 years significant persistence
no economically exploitable
persistence (no improvement
Sarnat 1972 efficient set approach 56 mutual funds 1946–1969 12 years 12 years
compared to the random
choice selection)
the composition of efficient
Levy & Sarnat
efficient set approach 100 mutual funds 1959–1980 11 years 11 years sets not better than that based
1984
on random choice
Appraisal Ratios, evidence of persistence but
Lehmann &
Jensen Alpha & 130 mutual funds 1968–1982 5 years 5 years results are sensitive to
Modest 1987
several APT alphas performance metrics
persistence when using MVRc,
Levy & Lerman
efficient set approach 100 mutual funds 1959–1980 11 years 1–11 years SSDRd, or TSDRe criteria
1988
(riskless asset included)
a
SP refers to selection period
b
HP refers to holding period
c
MVR refers to mean-variance efficiency criterion with riskless asset
d
SSDR refers to the second degree stochastic dominance efficiency criterion with riskless asset
e
TSDR refers to the third degree stochastic dominance efficiency criterion with riskless asset
Using the same data as Levy and Sarnat (1984) Levy and Lerman (1988) extend the work of the
formers to test the predictive power of investment decision criteria that use also information about the
riskless asset. While keeping the selection period always at the 11 years, Levy and Lerman vary the
length of the holding period from a maximum of eleven years for entry in 12th year to one year for the
ultimate year entry. Generally, the results indicate that there is a definite value to using ex post
information for ex ante portfolio selection, when the selection of efficient sets is based on mean-
variance criterion with riskless asset (MVR criterion), or the second or the third degree stochastic
dominance criterion with riskless asset (SSDR or TSDR criterion, respectively).
3. The Studies of the 1990s
Table 2 provides an overview of persistence studies of the 1990s and reveals one general trend in the
research design of the performance persistence studies; i.e., the shift to the use of shorter selection
period and holding period compared to those used in the earlier studies carried out in the time period
from 1960s to 1980s.
Christopherson and Turner (1991) classify managers according to the style and use a single
index reflecting that style instead of a broad market index in determining the manager alpha (named as
style index alpha). They conclude that the relationship between alpha over a previous three-year period
and an alpha in the subsequent one-, two- or three-year period does not exist.
Bogle (1992) ranks the annual raw returns of over 330 equity funds for 10 successive years for
the 1981-1990 time period. By comparing the average ranking of the TOP 20 funds for the former
5. 121 International Research Journal of Finance and Economics - Issue 34 (2009)
period to their average ranking for the subsequent period he finds no persistence in rankings from one
year to the next. In the interdecade return comparisons (1971-1980 vs. 1981-1990), the rankings are
even less meaningful
Grinblatt and Titman (1992) examine the performance persistence of 279 mutual funds over the
1975-1984 period using the methodology based on the eight-portfolio benchmark (P8)2 . 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 reveals positive persistence which cannot be explained by inefficiencies in
the benchmark that are related to firm size, dividend yield, past returns, skewness of return distribution,
interest rate sensitivity, or CAPM beta.
In another study Grinblatt and Titman (1993) develop a new innovative performance metrics
named as Portfolio Change Measure which evaluates performance on the basis of changes in quarterly
portfolio holdings of 155 funds for the time period 1975-84. The results show strong evidence of
persistence for the entire sample of funds and weaker evidence of persistence for subsamples of
aggressive growth, growth, and growth-income funds. Therefore, authors conclude that the observed
persistence in performance of the entire sample of funds is not due to consistent outperformance of
aggressive growth funds.
Using the survivorship bias-free sample of 41 nonmunicipal bond funds Blake et al. (1993)
examine whether past alphas are predictive of future alphas. They divide the 10-year period into two 5-
year subperiods and three 3-year subperiods (excluding the first year of data in this case). While all of
the models used by Blake et al. produce broadly similar ranking of funds, none of them is useful in
selecting funds that have higher alphas in subsequent periods. The authors analyze also the larger
samples in which the potential survivorship bias were not taken into account and find some evidence of
predictability.
Table 2: Performance persistence studies of the 1990s
Table 2 provides an overview of persistence studies of the 1990s and reveals one general trend in
the research design of the performance persistence studies; i.e., the prominent shift to the use of
shorter selection period and holding period compared to those employed in the earlier studies
carried out in the prior three decades (from 1960s to 1980s). The majority of studies find at least
some evidence of persistence which in most cases is explained by portfolio characteristics or/and
expense ratios.
Method for Type of funds Time Length of Length of
Authors performance and size of period selection holding Results
evaluation sample period period
Christophersson &
style index alphas 177 equity funds –1989 3 years 1–3 years no persistence
Turner 1991
equity funds from
330 (1981) to 829 1981–1990 1 year 1 year no persistence
Bogle 1992 total returns
(1990)
177 equity funds 1971–1990 10 years 10 years no persistence
Grinblatt & Titman
8-factor (P8) alpha 279 equity funds 1975–1984 5 years 5 years evidence of persistence
1992
several one-, 3-, 6- 41 non-municipal 1979–1988 5 years 5 years
Blake et al. 1993 no persistence
index alphas bond funds 1980–1988 3 years 3 years
persistence concentrated on inferior
Elton et al. 1993 3-index alpha 143 equity funds 1965–1984 10 years 10 years
performance
Portfolio Change evidence of persistence – weaker
Grinblatt & Titman
Measure (no 155 mutual funds 1975–1984 56 months 55 months evidence when style differences are
1993
benchmarks required) taken into account
3 months
total returns, Sharpe
Hendricks et al. 165 U.S. equity 6 months short-term persistence (particularly
Ratio, alphas based on 1974–1988 1 year
1993 growth funds 1 year among worst-performing funds)
various bechmarks
2 years
2
P8 was suggested by Grinblatt & Titman (1989). The basic idea behind the formation of this benchmark is that various
firm characteristics are correlated with their stocks’ factor loadings. As a result of this, portfolios constructed from stocks
classified by securities characteristics can be used as proxies for the factors. The P8 benchmark, formed from groupings
of the passive portfolios’ returns just described, consists of four size-based portfolios, three dividend-yield-based
portfolios, and the lowest past returns portfolio.
6. International Research Journal of Finance and Economics - Issue 34 (2009) 122
Method for Type of funds Time Length of Length of
Authors performance and size of period selection holding Results
evaluation sample period period
1 month 1 month
total returns
Goetzmann & 1–3 years 1–3 years evidence of persistence at its strongest
728 equity funds 1976–1988
Ibbotson 1994 1 month 1 months on the very short term
Jensen Alphas
2 years 2 years
total returns, Jensen
Alpha, Appraisal
equity funds from evidence of persistence within top-
Brown & Ratio, 3-index alpha,
372 (1976) to 829 1976–1988 1 year 1 year and worst-performers but also
Goetzmann 1995 3-index Appraisal
(1988) occasional reversals
Ratio, and
characteristic return
300 U.S. equity
1988–1993 3 years 3 years ambiguous results
total returns, Appraisal funds
Kahn & Rudd 1995 Ratio, and selection persistence in risk-adjusted
returns 195 bond funds 10/90–9/93 1 yr 5 mos 1 yr 5 mos performance, no persistence in total
returns
equity funds from
mixed results; strong persistence
220 (1971-1972)
Malkiel 1995 total returns 1971–1991 1 year 1 year during 1970s, no persistence during
to 684 (1990-
1980s
1991)
4-index alpha of Elton 188 U.S. equity evidence of persistence at its strongest
Elton et al. 1996 1977–1993 1–3 years 1–3 years
et al. (1996) funds using equal lengths of SP and HP
total returns strong evidence of persistence
270 common
Gruber 1996 4-index alpha of Elton 1985–1994 1–3 years 1–3 years particularly when 4-index alpha is
equity funds
et al. (1996) used as performance metric on SP a
abnormal returns
Volkman & Vohar 10/80– persistence over 1- to 3-year HPb
based on various 332 funds 1–5 years 1–4 years
1996 12/89 based on 3- and 4-year SP
models
total returns, Jensen short-term persistence in total returns
Alpha, Fama-French 1,892 U.S. equity explained by characteristics of
Carhart 1997 1962–1993 1–3 years 1–5 years
3-factor alpha, Carhart funds portfolio holdings and expense
4-factor alpha differences
1 year 1 year no reliable evidence of persistence
Phelps & Detzel several multi-index
87 mutual funds 1984–1994 2 years 2 years (occasional persistence observed is
1997 alphas
3 years 3 years explained by style differences)
total returns, Sharpe 1 year 1 year
Ratio, Treynor Ratio, U.S. equity funds 3 years 3 years weak evidence of persistence
Sauer 1997 Jensen Alpha, and the from 249 (1976) 1976–1992
explained by style differences
Elton et al. 3-factor to 1,365 (1992) 5 years 5 years
alpha
weak persistence that deteriorates
Detzel & Weigand characteristic-adjusted significantly after adjusting for beta,
61 equity funds 1975–1995 1 year 1 year
1998 returns expense ratios, firm size, and
investment style
93 mutual funds
no general persistence; inferior
with experienced
Porter & Trifts 1998 total returns 1986–1995 5 years 5 years performance persists particularly for
portfolio
funds with high expenses
managers
1 year
1 year evidence of persistence
total returns, group- 131 U.K. 2 years
Allen & Tan 1999 1989–1995
adjusted alpha investment trusts 1 month 1 month no evidence of persistence, but rather
6 months 6 months performance reversal
unconditional and
85 U.K. American
Fletcher 1999 conditional Jensen 1985–1996 1 year 1 year no evidence of persistence
unit trusts
Alphas
a
SP refers to selection period
b
HP refers to holding period
As a part of the informational efficiency study of managed portfolios Elton et al. (1993)
examine the persistence of the alphas of 143 mutual funds using the three-factor variant of the
conventional CAPM3. They rank the decile-portfolios from two successive decades and find highly
significant correlation between these two ranks. Furthermore, a regression of the three-factor alpha of
3
The factors employed by Elton, Gruber, Das, and Hlavka (EGDH 1993) are the return on the S&P 500 index, the return
on a non-S&P equity index that has been made orthogonal to the S&P index, and the return on a bond return index that
has been made orthogonal to both the S&P and the non-S&P equity index.
7. 123 International Research Journal of Finance and Economics - Issue 34 (2009)
the latter period on alpha of the earlier period is significant at the 5 percent level. However, the authors
are somewhat reserved in generalizing the results due to the strong persistence of inferior performance.
In one of the most widely-cited studies of the fund literature Hendricks et al. (1993) examine
performance persistence in a sample of open-end, no-load, growth oriented equity funds over the 1974-
1988 time-period. The authors launch the already-established concept of “hot-hands” (which appears also
in the title of this study) to describe the short-term nature of performance persistence; the results show that
funds that outperform in the most recent year continue to outperform in the near term peaking at the
holding period of the same length. Furthermore, funds that perform poorly during the most recent one-year
period tend to underperform also in the near future. According to the results, the persistence of inferior
performance is even stronger than persistence of superior performance. Moreover, Hendricks et al. (1993)
prove that the results are robust on several potential biases (i.e., benchmark inefficiency, spurious
persistence, nonlinearities between fund returns and benchmark returns, time-varying betas and data-
snooping bias introduced by Lo and MacKinlay 1990).
Goetzmann and Ibbotson (1994) analyze monthly total returns of 728 mutual funds over 13-year
period (1976-1988). Using total returns and the Jensen alphas as performance measures they examine the
power of various lengths of selection periods to predict the performance measured from holding periods of
the same length. The time horizons tested in this study are one year, two and three years and one month.
Generally, the results are significant, i.e., past performance has some predictive power on future
performance for all time horizons tested.
To test robustness of the results over the conjecture whether the performance persistence is
related more to investment style than skill, Goetzmann and Ibbotson (1994) perform the same tests on
a sub-sample that consists only of the relatively homogenous growth funds. The tests indicate that the
performance persistence is not likely to be due to style differences. The study of Goetzmann and
Ibbotson (1994) is innovative in the sense that for the first time in the mutual fund literature it controls
for momentum effect; In order to discriminate whether the one-month persistence is due to momentum
effect or a long-term phenomenon (related possibly to risk level), Goetzmann and Ibbotson perform a
randomization test, which explicitly uses the long-term mean return to the fund as the control to test
whether the preceding month return has any additional explanatory power. They find the preceding
month’s ranking to have power to predict the next month’s ranking above and beyond the effects
caused by differences in long-term means.
Brown and Goetzmann (1995) examine to what extent the previous-year performance of a fund
can predict the performance of successive year over the 1976-1988 period. The authors use several
alternative performance measures4 and find clear evidence of relative performance persistence but
instead, evidence of absolute persistence is weaker and dependent of the time period being evaluated.
Most of the persistence phenomenon observed is due to consistent underperformance rather than due to
consistent outperformance. In this respect the results are parallel to those of Jensen (1968), Shukla and
Trzcinka (1994), Carhart (1997), Lunde et al. (1999), Teo and Woo (2001) and Fletcher and Forbes
(2002). In another respect, the findings of Brown and Goetzmann are parallel to those of Malkiel
(1995), who ⎯ using quarterly data of equity mutual funds from 21-year period from 1971 to 1991 ⎯
tests the prediction power of the previous-year return of a fund on the corresponding return of
successive year. He finds considerable persistence in fund returns during the 1970s, but no consistency
of them during the 1980s. Similar results are also reported later by Droms and Walker (2001a) who
hypothesize that time period dependency may be due to the size anomaly; small-cap stocks
outperformed the S&P 500 during the 1970s, while reverse was true for the 1980s. The results of
Detzel and Weigand (1998) reveal that besides size anomaly, also style characteristics of the stocks
held by equity funds explain the persistence findings for the 1976-1985 period; allowing for the
market-cap of the stocks included in funds’ portfolios and manager investment styles as additional
explanatory variables, all of the persistence in fund performance disappears. However, the explanatory
4
The tests are done using the total return, the Jensen Alpha, the Treynor & Black (1973) Appraisal Ratio, the Three-index
alpha and the corresponding Appraisal Ratio, and “group-adjusted” return (the raw return minus the return for the fund
style.
8. International Research Journal of Finance and Economics - Issue 34 (2009) 124
power of size anomaly is not unambiguous; Both Quigley and Sinquefield (2000), using U.K. fund
data, and Davis (2001), using U.S. fund data, report persistence explained by the worst-performing
small-cap funds.
Kahn and Rudd (1995) examine performance persistence of both equity and fixed income funds
analyzing them separately. For equity funds, a selection period of three years is used to predict the
performance of holding period of the same length, while for fixed income funds the corresponding
length of periods is a year and five months for both selection and holding periods. In the case of equity
funds, regression analysis finds evidence of persistence at the 5 % level only for Appraisal Ratios.
Using the contingency tables approach none of the tests for three performance measures shows
evidence of persistence. In the case of fixed income funds, both regression and contingency table
analyses show significant persistence of both style-adjusted returns and Appraisal Ratios.
Elton et al. (1996) examine the survivorship bias-free sample of common stock funds followed
from 1977 to the end of 1993. They extend the three-index model of Elton, Gruber, Das and Hlavka
(EGDH 1993) by introducing one more index to account for the performance of growth versus value
stocks. Furthermore, Elton et al. (1996) refine the EGDH model using differential returns in measuring
size (i.e., differential return between a portfolio of small stocks and large stocks) and types of stocks
(i.e., differential return between a portfolio of growth stocks and a portfolio of value stocks) as factors
besides the return on the S&P 500 index and the bond index return. They form decile portfolios of
funds based on four measures (i.e., total returns, one- and three-year four-index alpha and the t-statistic
of the four-index alpha) and observe how the decile portfolios perform in follow-up period whose
performance is measured with one- and three-year four-index alpha. For three-year holding period, any
other ranking criteria studied, except for total return, leads to a significant rank correlation. The same
analysis is repeated with one-year holding period. In this case, ranking techniques involving one year
of past data generally perform much better than those involving 3 years of past data. Similarly to the
results of Hendricks et al. (1993), the fraction portfolios formed on the basis of total return are highly
correlated with future alpha when alpha is measured over a one-year period, but the relationship
deteriorates when future alpha is measured over three years. However, when ranking is done on a risk-
adjusted basis the predictability increases as performance is measured over the longer (three-year)
period.
Using raw returns and the four-index alpha of Elton et al. (1996) as performance measures,
Gruber (1996) studies the survivorship bias-free sample of common stock mutual funds over the 1985-
1994 period. At the end of the each year, funds are ranked and placed to the decile portfolios on the
basis of a particular selection criterion. Gruber finds strong performance persistence with both one- and
three year horizons and also the four-index alpha’s superiority to forecast future performance
determined on the basis of either risk-adjusted or raw returns.
Volkman and Wohar (1996) analyze the performance persistence of 112 mutual funds over the
1980-1989 period. They use three different empirical models to test the performance persistence in
relation to each of 20 combinations of selection periods of 12, 24, 36, 48 and 60 months, and holding
periods of 12, 24, 36 and 48 months. All three models show persistence in abnormal returns over a
two- to three-year holding period based on a three- to four-year selection period.
In one of the most oft-cited studies in the mutual fund literature, Carhart (1997) examines the
survivorship-bias free data consisting of monthly returns of diversified equity funds over the 1962-
1993 period. Replicating the methodology of Hendricks et al. (1993), he forms decile portfolios of
mutual funds on lagged one-year returns and estimates performance on the resulting portfolios. Though
the results of Carhart strongly support the short-term performance persistence he notes that most of the
short-term persistence observed is explained by common factor sensitivities of his four-factor model5,
and differences in expenses and transaction costs.
5
Carhart (1997) constructs his 4-factor model by including on the Fama & French 3-factor model an additional factor
capturing Jegadeesh and Titman’s (1993) one-year momentum anomaly. This is motivated by the 3-factor model’s
inability to explain cross-sectional variation in momentum-sorted portfolios (documented by Fama & French 1996).
9. 125 International Research Journal of Finance and Economics - Issue 34 (2009)
Using selection periods of one year, three and five years and investment periods of equivalent
length Sauer (1997) finds statistically significant performance persistence in all horizons studied for
U.S. equity funds over the 1976-1992 period. Taking account of sporadic evidence of reversion in
relative fund performance during some successive years, the shorter the horizons the stronger
persistence. The same causality is also found for the zero-investment long/short octile portfolios
formed on the basis of the 3-index EGDH alpha. In addition, Sauer examines persistence separately for
the growth and growth and income mutual funds, respectively. When the full sample is partitioned by
investment objective, the statistically significant persistence in mutual fund performance is no longer
evident on five-year horizons. Unfortunately, Sauer does not report the corresponding results with
shorter horizons.
An interesting contribution to persistence literature is provided by Phelps and Detzel (1997)
who mimic the study of Goetzmann and Ibbotson (1994) by examining the predicting power of past
alphas for the same performance measures for the subsequent period of equal length. Based on several
empirical tests with the multi-index models with varied number of factors the authors argue that the
positive persistence documented in several studies is the result of persistence in broad equity classes
(macropersistence) rather than sustainable managerial ability (micropersistence). In other words, the
observed persistence would result from factors that a generic index as a surrogate for market return
cannot adequately capture. Also, according to Detzel and Weigand (1998), fund performance
corresponds to the performance trends of the size and style classes in which funds invest. Employing
the model suggested by Daniel and Titman (1997) that directly relates mutual fund returns to the
characteristic of the stocks held by funds, the authors find that the adjustment of fund returns for both
the size of the firms in which funds invest, and for financial ratios intended to capture fund manager
investment styles explains all the persistence in mutual fund performance.
Porter and Trifts (1998) examine the performance of 93 experienced fund managers over the
ten-year period of 1986-1995 using relative percentile ranks based on quarterly compounded, annual
total returns measured against funds with the same investment objective. The results show that for the
experienced managers studied, superior performance in one five-year period is not predictive of
superior performance over the next five years. However, inferior performance persists particularly for
funds with above average expense ratios.
Allen and Tan (1999) investigate the performance persistence of 131 U.K. investment trust
company managers over the 1989-1995 period. The authors examine the prediction ability of both raw
returns and that of style-index alphas for the one-year, half-year and monthly periods. According to the
results, prior one-year performance includes definite information about future performance for the
periods of both one year and two years on the basis of both measures. By contrast, for shorter periods
the results support performance reversal rather than persistence.
Fletcher (1999) examines the performance of a sample of 85 UK American unit trusts using
both the unconditional Jensen alpha and the conditional Jensen alpha (developed by Ferson and Schadt
(1996)) as follows: At the beginning of each year all trusts are ranked on the basis of their cumulative
excess returns over the previous year and grouped into quartile portfolios. Equally weighted monthly
excess returns are then estimated over the next year. Fletcher finds no evidence of the persistence in
performance for this sample of trusts.
4. The Studies of the 2000s
Table 3 provides a summary of persistence studies published heretofore in the new millennium.
Compared to the studies published in the previous decade the average length of both selection period
and holding period has decreased. Thus, the long-term tendency in the persistence literature towards
using shorter past data to predict future performance for shorter holding periods has continued also in
2000s.
Blake and Morey (2000) compares the Morningstar ratings as a predictor of mutual fund
performance to the established performance metrics (i.e., total returns, Sharpe Ratio, Jensen alpha, and
10. International Research Journal of Finance and Economics - Issue 34 (2009) 126
4-index alpha of Elton et al. (1996)). Based on two sample groups for time periods of different length
the comparison indicates that the Morningstar ratings are in the middle in terms of predicting future
performance. For the longer sample period based on 10-year selection period, total returns and the 4-
index alpha do worse, but the Sharpe Ratio does considerably better than the Morningstar ratings. For
the shorter sample period based on 3-year selection period the results show somewhat surprisingly that
Morningstar ratings predict the future performance significantly better than the above-mentioned
established performance metrics. However, after controlling for the fact that for the majority of funds
in the sub-sample employed the Morningstar stars are based on up to 10 years of return data in contrast
with 3-year selection period of performance metrics being compared, the superior ability of the
Morningstar method disappears. Thus, in contrast to the prevailing trend of the persistence literature,
this finding of Blake and Morey (2000) would indicate that it could be still worthwhile to use return
history older than 3 years for the purposes of predicting future performance. The authors conclude that
the Morningstar rating system is able to "identify" low-performing funds since funds with less than
three stars generally have much worse future performance than other groups. Instead, only weak
evidence that the five-star (highest-rated) funds would outperform the four- and three-star funds is
found. Thus, the Morningstar rating system, like the other established performance metrics, seems to
be more capable in identifying inferior than superior performers of the future due to the persistence in
poor performance.
As a part of the larger study of the value of active mutual fund management Chen et al. (2000)
investigate performance persistence by examining the performance of both the holdings and the trades
of mutual funds for the 1975-1994 period. Controlling for differences in stock characteristics, the
results generally do not support the persistence of fund performance, although persistence in
unadjusted returns on mutual fund portfolio holdings exist.
Dahlquist et al. (2000) estimate performance persistence of Swedish mutual funds by treating
previous-year alphas obtained from various regressions as an attribute of future success. The results
show persistence neither for equity nor bond funds, but among money market funds it does exist. Using
monthly returns of all U.K. equity funds for the 1978-1997 period Quigley and Sinquefield (2000) find
evidence of persistence among inferior performers but no persistence among superior performers.
Contrary to size anomaly, persistent underperformance is concentrated on small-cap funds.
Jain and Wu (2000) examine 117 mutual funds that were advertised from July 1994 through
June 1996 in Barron’s or Money magazine by comparing pre- and post-advertisement performance of
these funds. Using four different performance measures6 they find that advertised funds have superior
performance prior to advertisement year, but turn to underperformers in the year following advertising.
6
Jain and Wu (2000) employ excess return over return on funds with the same investment objective (noted as the similar-
funds-adjusted return), and that over return on S & P 500 index (noted as the S&P 500-adjusted return), the Jensen alpha
and the Carhart 4-factor alpha as performance metrics.
11. 127 International Research Journal of Finance and Economics - Issue 34 (2009)
Table 3: Performance persistence studies of the 2000s
Table 3 provides a summary of persistence studies published heretofore in the new millennium.
Compared to the studies published in the previous decade the average length of both selection
period and holding period has decreased. The majority of the studies of the 2000s find evidence of
persistence but many of them with provisions. Evidence of persistence at least among worst-
performing funds can be considered noteworthy. However, overall results are somewhat mixed
varying from strong persistence to reversal depending on performance metrics, observation period
and the sample data employed.
Method for Type of funds and Time Length of Length of
Authors performance size of sample period selection holding Results
evaluation period period
1 year evidence of persistence
3 years particularly in inferior
263 U.S. equity funds 1983–1997 10 years
Morningstar ratings, performance (the best
5 years
Blake & total returns, Sharpe predictor: Sharpe Ratio)
Morey Ratio, Jensen Alpha, 4- 1 year evidence of persistence
2000 index alpha of Elton et 3 years particularly in inferior
al. (1996) 635 U.S. equity funds 1990–1997 3 years performance (the best
5 years predictor: Morningstar
rating)
no general persistence using
total returns, U.S. mutual funds characteristic-based alphas;
Chen et al.
characteristic-based from 393 (1975) to 1975–1994 1 year 1 year persistence using unadjusted
2000
alpha of DGTWa (1997) 2,424 (1994) returns explained by
momentum effect
unconditional and 210 Swedish funds robust persistence among
Dahlquist conditional 2-index (126 equity funds, 42 money market funds, no
1993–1997 1 year 1 year
et al. 2000 alphas for equity and bond funds, 42 money persistence among other
bond funds market funds) funds
similar-funds-adjusted
Jain & Wu return, S&P 500- 117 mutual funds 7/1993– no persistence among
1 year 1 year
2000 adjusted return, Jensen (recently advertised) 6/1997 recently advertised funds
Alpha, 4-factor alpha
73 non-conventional persistence limited to the
1 year 1 year
bond funds (high- high-yield bond funds
Philpot et
Sharpe Ratio yield, global, and 1988–1997
al. 2000
convertible bond 5 years 5 years no persistence
funds)
Quigley & 1 year 1 year persistence only among
total returns, Fama- 311 U.K. equity unit
Sinquefiel 1978–1997 worst-performing small-cap
French 3-factor alpha trusts (on average) 3 years 3 years
d 2000 funds
weak evidence of short-term
persistence among the best-
Davis Fama-French 3-factor
4,686 equity funds 1962–1988 3 years 1 year performing growth funds
2001 alpha
and among the worst-
performing small-cap funds
raw returns
Droms & 10 years 10 years no long-term persistence
Jensen Alphas
Walker 151 U.S. equity funds 1971–1990
2001a 1-3 yrs
raw returns 1 year short-term persistence
ahead
Droms & International equity performance persists over 1-
Walker raw returns funds from 11 (1977) 1977–1996 1 year 1-4 years year holding period but not
2001b to 473 (1996) over longer holding periods
1 year 1 year short-term persistence
total returns growth and income
particularly in growth fund
equity funds from 3 years 3 years
returns
ter Horst Jensen Alpha sample of 2,678 U.S.
1989–1994 evidence of persistence
et. al. 2001 equity funds (number
of funds within fund among worst-performing
3 years 3 years
Carhart 4-factor alpha classes not reported) funds (esp. among income
equity funds)
1 year 1 year persistence in total and
Group-adjusted returns,
Carhart et 2,071 diversified 5 years 5 years group-adjusted returns
total returns 1962–1995
al. 2002 equity funds deteriorating after end-of-
3 years 3 years
4-factor alpha sample or look-ahead
12. International Research Journal of Finance and Economics - Issue 34 (2009) 128
Method for Type of funds and Time Length of Length of
Authors performance size of sample period selection holding Results
evaluation period period
conditioning
no evidence of persistence
Detzler 4-factor alpha of
423 mutual funds 1990–1996 3 years 1 year among publicly-ranked
2002 Detzler (2002)
funds
Carhart 4-factor alpha ambiguous; (performance
Jensen Alpha reversal based on
Fletcher & APT alpha U.K. equity trusts conditional alpha –
Forbes from 139 (1982) to 1982–1996 1 year 1 year persistence based on
2002 724 (1996) unconditional alphas but no
Conditional alpha persistence based on Carhart
alpha)
relative excess returns
evidence for short-term
(over the equally- South African general
Collinet & 6 months 6 months persistence (particularly for
weighted average return equity unit trusts from 1980–1999
Firer 2003 the 1995–1999 period)
of all the funds) 7 (1980) to 43 (1998)
Sharpe ratios 3 years 3 years medium term persistence
persistence for most equity
Jan & 16,345 funds of all
efficient set approach 1961–2000 1 year 1 year and money market funds;
Hung 2003 type
reversal for most bond funds
5-factor model Canadian equity funds short-term persistence; some
Deaves 1-5 years
conditional CAPM from 110 (1988) to 1988–1998 1 year evidence for medium-term
2004 ahead
alpha 300 (1998) persistence
Combinatio
n of 1-yr &
Jan & 3,316 U.S. equity short- and medium-term
Carhart 4-factor alpha 1961–2000 3-yr 1 year
Hung 2004 funds persistence
rankings
(lagged)
equity funds from
Prather et
multi-factor alpha 2,124 (1996) to 3,391 1996–2000 1 year 1 year no persistence
al. 2004
(1999)
Bollen & 230 U.S. equity funds
Busse Carhart 4-factor alpha (new funds after 1985 1985–1995 3 months 3 months very short-term persistence
2005 not added)
Morningstar
ratings/Sharpe Ratio, no persistence among the
Morey Jensen Alpha, 4-factor 4/1987– funds upgraded for the first
273 U.S. equity funds 3 years 3 years
2005 alphas of both Elton et 6/2000 time to five-star funds by
al. (1996) and Carhart Morningstar
(1997)
various multi-factor 1 year
Busse & 230 U.S. equity funds short-term persistence
alphas (Bayesian) for 3 months
Irvine (new funds after 1985 1985–1995 3 months particularly by using annual
SP; standard multi-
2006 not added) 3 years selection period
factor alphas for HP
Harlow & U.S. equity funds 1 month strong persistence for 1-
Fama-French 3-factor
Brown from 131 (1981) to 1979–2003 3 years 3 months month and 3-month holding
alpha
2006 5,614 (2003) 1 year periods
1 year persistence among growth-
U.S. equity funds
Kosowski oriented funds; non-
Carhart 4-factor alpha from 231 (1971-1975) 1975–2002 1 year
et al. 2006 3 years persistence among income-
to 1,788 (1975-2002)
oriented funds
Polwitoon raw returns 1 year mixed results; persistence
global bond funds
& for some consecutive years –
from 103 (2003) to 1993–2004 1 year
Tawatnunt Sharpe Ratios 3 years reversal for some other
183 (1997)
achai 2006 consecutive years
short-term performance
Huij & U.S. equity funds
Carhart 4-factor alpha persists but varies across
Verbeek from 362 (1984) to 1984-2003 1 year 1 month
(Bayesian) styles (strongest for small
2007 4,973 (2003)
cap/growth funds)
a
DGTW refers to the method introduced by Daniel, Grinblatt, Titman & Wermers (1997)
13. 129 International Research Journal of Finance and Economics - Issue 34 (2009)
Philpot et al. (2000) analyze performance persistence of 73 non-conventional bond funds (high-
yield bonds, global issues and convertible bonds) for the 1988-1997 time period and find evidence of
short-term performance persistence for the high-yield bond fund sub-sample, but no persistence for the
general class or for other classes of funds. The persistence found on the basis of one-year Sharpe
Ratios disappears, as the selection period is extended to five years and the sample period examined is
divided into two sub-periods of the equal length.
Davis (2001) examines the relationship between equity fund performance and manager style by
employing the Fama-French (1993) alpha as performance metrics. Particularly, Davis addresses
whether any particular investment style reliably delivers abnormal performance and furthermore,
whether any evidence of performance persistence can be found when funds with similar styles are
compared. Davis does not find positive abnormal returns over the 1965-1998 period although he does
find some evidence of short-term performance persistence among best-performing growth funds.
However, this persistence is not sustained beyond one year.
The study of Droms and Walker (2001a) follows the methodology developed by Goetzmann
and Ibbotson (1994), Brown and Goetzmann (1995), and Malkiel (1995) to test for performance
persistence among equity mutual funds over the two decades from 1971 to 1990. The results show no
long-term persistence based on either total returns or the Jensen Alphas. Instead, evidence of short-
term persistence is found for periods of one, two and three years. Consistent with the findings of
Brown and Goetzmann (1995), and Malkiel (1995), the persistence is more pronounced during the first
decade of the 1970s than the 1980s.
Droms and Walker (2001b) follows the same type of methodology also on another study that tests
for short-term performance persistence in international equity mutual funds over the 20-year period from
1977 to 1996. Using annual returns as performance measures, Droms and Walker (2001b) find statistically
significant performance persistence for 1-year holding periods, but no persistence for 2-, 3- or 4-year
periods. The similar conclusions are also drawn by ter Horst et al. (2001) who examine the impacts of
survivorship bias and look-ahead bias with the sample of U.S. growth and income equity funds for the
1989-1994 period. For 3-year selection period and holding period the results show evidence of risk-adjusted
performance persistence only among worst-performing funds (particularly among income equity funds).
Without any risk-adjustment procedures the same analysis shows no signs of medium-term persistence.
In a comprehensive study of selection bias issues in the context of mutual fund research Carhart
et al. (2002) find persistence in the performance of U.S. mutual funds. Employing three different
performance metrics7 the authors undertake Hendricks, Patel and Zeckhauser’s (1997) test for spurious
persistence due to survivorship8 and find the results to be robust to survivorship bias.
Using the sample consisting of 757 funds Detzler (2002) examines the performance of an
investment strategy based on mutual fund rankings by the popular press (Barron’s, Business Week and
Forbes). The results show that rankings correspond to higher returns 3, 6, and 12 months before the
publication dates of rankings, but the funds do not have superior performance in the post-ranking
periods of equal lengths. Furthermore, the ranked funds have often higher risk than their non-ranked
peers in both the pre-ranking and post-ranking periods, suggesting that funds receiving rankings may
also be risk-takers. The 4-factor alpha9 shows that the funds with rankings have higher risk-adjusted
performance during the pre-ranking period and negative performance in the post-ranking period
providing evidence against persistence. Thus, the results are very much consistent with the findings of
Jain and Wu (2000).
7
The three performance measures used by Carhart et al. (2002) are “group-adjusted” returns, the Jensen Alpha, and the
Carhart 4-factor model.
8
Hendricks et al. (1997) show that when performance is categorized finely, the relation between pre- and post-period
rankings will be J-shaped in a survivor-biased sample or using a look-ahead biased methodology. They devise a
regression test for this convexity, which Carhart et al. (2002) employ in their survivorship- and look-ahead -biased
samples.
9
The Detzler 4-factor alpha is based on following indices: S&P 500 index, the MSCI EAFE index, a small-cap index, and
the Lehman Brothers Aggregate Bond index.
14. International Research Journal of Finance and Economics - Issue 34 (2009) 130
Fletcher and Forbes (2002) find evidence of persistence in UK unit trust performance when
performance is determined by means of factor models based on the CAPM or APT. However, when
performance is estimated relative to the Carhart 4-factor model, the persistence disappears.
Interestingly, the use of the conditional performance measure developed by Ferson and Schadt (1996)
turns the observed persistence into significant reversal. Thus, Fletcher and Forbes (2002) conclude that
the persistence in performance of UK trusts is not a manifestation of superior stock selection strategy,
but can be explained by factors that are known to capture cross-sectional differences in stock returns.
Collinet and Firer (2003) analyze the relative performance of South African general equity unit
trusts from 1980 to 1999 using the relative excess returns (over the equally-weighted mean return of all
the funds in existence during the period) as a performance measure. The authors find evidence of
persistence when the selection of funds for 6-month holding period is based on performance from the
selection period of 6-12 months. According to the results, persistence is particularly evident during the
1995–1999 period using 6-month selection period. However, even within this period, there are cases
where rankings from one holding period to the next are random and also cases of reversed rankings.
Furthermore, the results of tests with longer holding periods are less conclusive; although strong
persistence is found over certain periods, the results are very sensitive to variations in both the ending
date of the selected sample period and the time period studied.
As a part of the larger study of mutual fund attributes and performance Jan and Hung (2003)
examine performance persistence of U.S. mutual funds over the 1961-2000 period. Forming winner
and loser portfolios based on one-year raw returns and testing the efficiency of these portfolios of
funds the authors find that persistence exists among 13 out of 24 fund categories examined. On the
other hand, evidence of performance reversal is found among 7 fund categories. According to the
results, persistence is more common among equity funds while reversal is typical in most bond fund
categories.
Another study of Jan and Hung (2004), using the same time period but somewhat smaller
sample of the same database, hypothesize that if mutual fund performance persists in the short run, it
should also persist in the long run. A division of the funds in the database on the basis of past 4-factor
alpha of Carhart (1997) – funds with strong past short-run and long-run performance rated as best –
reveals that in the subsequent year the best funds significantly outperforms the worst funds. The
authors conclude that mutual fund investors can likely benefit from selecting funds on the basis of not
only past short-run performance but also past long-run performance.
Deaves (2004) examines performance persistence of Canadian equity funds on the basis of
several performance measures. Using carefully constructed bias-free sample for the 1988-1998 period
he finds evidence of short-term persistence at its strongest when one-year selection period is used to
predict next year’s performance.
Prather et al. (2004) analyze the impact of numerous fund-specific characteristic on
performance of equity funds. The analysis includes 25 individual fund factors or characteristics within
the four broad categories of popularity, growth, cost and management. For the 1996-2000 period, they
find no evidence of persistence, but instead, a reversal pattern in mutual fund performance.
Studying daily returns of 230 U.S. equity funds from the 1985-1995 period, Bollen and Busse
(2005) find that the top decile funds managers generates statistically significant quarterly abnormal
returns that persist for the following quarter. The results are robust across stock selection, market
timing, and mixed strategy models, which suggests that misspecification of the performance model is
not the reason for evidence of persistence. However, the authors note that the economic significance of
the post-ranking abnormal returns is questionable given the transaction costs and taxes levied on a
strategy capturing the persistent abnormal returns of the top decile.
Morey (2005) examines the performance persistence of U.S. equity funds that have just
received their first 5-star rating from Morningstar. During the 3-year period following the rating
upgrade performance deteriorates dramatically in spite of the performance metrics used in evaluating
performance of holding period. In this sense, the results are parallel to those of Detzler (2002).
15. 131 International Research Journal of Finance and Economics - Issue 34 (2009)
Morey’s results are also robust across different sub-samples of funds (i.e., samples of actively managed
funds and growth funds).
Using a 3 × 3 classification system similar to that of Morningstar Harlow and Brown (2006)
sort the fund universe based on alphas of the Fama & French 3-factor model (1992, 1993), and
examine performance persistence both within these fund classes and in the aggregate level. Based on
the three-year selection period, the results indicate a strong degree of performance persistence in the
active U.S. equity fund sample for holding periods up to one year. The authors state that persistence is
particularly strong and highly statistically significant in the near short-term, i.e. for time periods of one
month and three months.
Applying a new bootstrap technique to the monthly net returns of the universe of U.S. equity funds
during the 1975-2002 period, Kosowski et al. (2006) find strong evidence of superior performance and
performance persistence among growth-oriented funds, but no corresponding evidence of income-oriented
funds. They rank funds using the unconditional four-factor alpha measured over one and three years prior to
one-year holding period.
Polwitoon and Tawatnuntachai (2006) examine performance persistence of US-based global
bond funds during the period of 1993–2004. Following the methodology of Elton et al. (1996) funds
are ranked on the basis of 1- and 3-year raw returns and 1- and 3-year Sharpe Rratios prior to
subsequent 1-year holding period. The results show that persistence is stronger using shorter selection
period, i.e., 1 year instead of 3 years. Although some evidence of performance persistence among
global bond funds is found, and the rank correlation is significant for all years, it is negative in 5 out of
11 years, indicating performance reversal almost as often as it indicates persistence.
Recently, several scholars have used Bayesian alphas as a performance measure (e.g., see Baks
et al., 2001; Pástor and Stambaugh, 2002a, 2002b; Bollen and Busse, 2005; Busse and Irvine, 2006;
Huij and Verbeek, 2007). The basic idea of the Bayesian approach is to include prior information
related to such issues as funds’ expenses, investors’ beliefs about managerial skills, benchmark pricing
abilities, or the returns on other mutual funds and benchmark factors, in the resulting estimates. Such
an approach can be motivated both by cross-sectional learning of investors (as noted by Jones and
Shanken 2005) and on the basis of statistical arguments only. The results of the studies applying the
Bayesian approach are promising since the superior prediction power of Bayesian alphas over standard
OLS alphas is documented most often.
Using daily returns of 230 U.S. equity funds and the Bayesian approach suggested by Pástor and
Stambaugh (2002b)10, Busse and Irvine (2006) compare the performance predictability of Bayesian alphas
with standard frequentist measures. When the returns on passive nonbenchmark assets are correlated with
fund holdings, incorporating histories of these returns in a Bayesian framework produces alphas that predict
future performance better than standard alphas do. During the 1985-1995 period being evaluated,
persistence is at its strongest when the Bayesian alphas estimated over one-year ranking period are used to
predict subsequent standard quarterly alphas. Also, the other selection periods tested (i.e., one quarter and
three years) show evidence of prediction power. Of Bayesian alphas based on various performance models
the best is that of the Carhart 4-factor model.
However, the predictive accuracy of Bayesian alphas is in most studies greatly affected by the
investor’s prior belief about managerial skill. Huij and Verbeek (2007) apply the Bayesian approach so
that it does not require investors to explicitly formulate their beliefs about managerial skill (i.e. the
prior), or to make assumptions about cross-sectional characteristics that drive performance. This is
done by incorporating the large cross-section of mutual fund alphas in measuring the skill of an
individual fund manager. The basic principle is to allow the prior to learn across other funds included
in the sample, in which case the resulting belief in managerial skill is no longer fully subjective, but
instead, it is entirely based on sample-period data. Using monthly return data of more than 6,400 U.S.
equity mutual funds Huij and Verbeek investigate short-run performance persistence over the period
10
Pástor & Stambaugh (2002b) show that the precision of estimates of fund performance could be improved by
incorporating a long time series of passive asset returns using Bayesian approach. Thus, mutual fund performance
measures need not be restricted to information on fund and passive assets over the life of the fund.
16. International Research Journal of Finance and Economics - Issue 34 (2009) 132
1984–2003. They find that when funds are sorted into decile portfolios based on 12-month ranking
periods, the top decile of funds earns a statistically significant, abnormal return of 0.26 percent in the
first month after ranking. This effect is robust to load fees that are involved with a strategy of chasing
winners. Furthermore, their results show that persistence varies across investment styles and it is
mainly concentrated in relatively young, small cap/growth funds.
5. Concluding Remarks
The preceding review of performance persistence literature reveals that plenty of studies have been
published both for and against the prediction power of past performance. The results of previous
studies also indicate that there is not only one truth on this issue. Firstly, as shown in several studies,
even contrary conclusions may sometimes be drawn by using the same sample but different
methodology of performance evaluation (e.g., see Kahn and Rudd, 1995; Fletcher and Forbes, 2002).
Some methodologies seem to be more sensitive than some others to identify performance
persistence. For example, comparing performance differences between quantile portfolios may result in
contrary conclusions than employing the rank correlation test for the same sample.
Another source of bias that may affect the inferences on performance persistence stems from
performance model employed. In most cases, when the performance model takes account of
differences in portfolio characteristics, the evidence of persistence usually deteriorates, and in some
cases vanishes completely. Adding other factors such as size, book-to-market, or momentum besides
general market factor into the performance model may change the results drastically. E.g., the results of
Carhart (1997) show that evidence of persistence may be explained by the omitting momentum factor.
The above-described bias is explained by differences in investment styles of fund managers. For
example, in the second half of 1990s many funds followed either value or growth strategy. Had style
bias not been taken into account in the performance model, the chances that a value-oriented fund
would have outperformed a growth-oriented fund were very low. Correspondingly, in the beginning of
the ongoing millennium the case has been contrary. Unfortunately, style bias cannot be completely
circumvented by employing performance metrics (such as the Sharpe Ratio, for example) that are not
based any benchmarks. Pätäri (2008) compares an extensive set of performance metrics that are based
on both full-scale and partial-scale measures of risk (i.e., measures of downside risk) derived from a
portfolio’s own return distribution without using any benchmarks. The results show that due to the
asymmetries of return distributions the relative performance of funds depends on a risk measure
employed. It is highly probable that the sensitivity of total-risk based performance rankings to the
selection of a risk measure is a reflection of style bias.
Style bias has been tried to alleviate by using style-adjusted performance metrics but even that
approach can not protect from another source of bias. While style bias stems from performance metrics
employed, a misclassification bias is caused by a fund’s deflection from its stated investment policy.
Several studies have documented severe and frequent divergences between the actual and stated
investment policies of mutual funds (e.g., see diBartolomeo and Witkowski, 1997; Brown and
Goetzmann, 1997; Kim et al., 2000; Castellanos and Alonso, 2005; Detzel, 2006). According to the
results the average divergence rate ranges from 33 per cent to as high as 50 per cent within some fund
categories. Therefore, the fact that very many mutual funds are benchmarked against irrelevant factors
may induce spurious persistence.
As noted by several scholars, performance persistence studies are prone to several biases that
stems from ex-post conditioning of data. The most well-known of these is survivorship bias that stems
from including only the funds that exist at the end of sample period. Though the survivorship bias is
quite often offered as an explanation for the results supporting performance persistence the opinions on
the degree of the impact of survivorship bias on the results of persistence studies vary strongly among
scholars (e.g., compare the views of Grinblatt and Titman, 1989; Hendricks et al., 1993; Wermers,
1997, and/or Sauer (1997) to those of Malkiel 1995; Gruber, 1996; ter Horst et al., 2001; Carhart et al.,
2002, and/or Deaves, 2004). According to some studies, persistence is even stronger in full samples
17. 133 International Research Journal of Finance and Economics - Issue 34 (2009)
than survivor-only samples (e.g., see Hendricks et al., 1993; Carpenter and Lynch, 1999; Carhart et al.,
2002) while some other studies concludes that survivorship bias may in certain conditions induce
performance reversal rather than persistence (e.g., see Brown et al., 1992; Grinblatt and Titman, 1992).
Another form of data-conditioning stems from look-ahead bias, which is inherent any test of
performance persistence. A common methodology in performance persistence studies is to rank funds
and assign them to fraction portfolios on the basis of their performance from the preceding selection
period. Look-ahead bias arises because funds disappear in non-random way during the selection period
or holding period, i.e., the attrition rate of funds within fraction portfolios is not stable. Thus, an
essential approach to control look-ahead bias is to model the survival process of funds, and secondly,
to analyze how it relates to their past performance. Though this approach is followed very seldom in
mutual fund persistence studies the recent studies indicate that look-ahead bias is not very severe in
samples of mutual funds if survivorship bias is controlled (e.g. see ter Horst et al., 2001; Carhart et al.,
2002; Deaves, 2004).
The third form of data-conditioning bias called a self-selection bias is caused by the voluntary
nature of data provision. It exists in mutual fund research mainly because underperforming funds do
not necessarily send their records to data vendors. A self-selection bias may also occur in the context of
fund mergers when a fund management company launches two funds at year-end, and decides to merge
the underperformed fund with the outperformed fund at the end of the next year. When there is
typically 12-month delay before a fund’s records are sent to the administrator of mutual fund database
the company may be tempted to provide the full record of the outperformed fund while omitting the
data of the underperformed fund. It is therefore likely that companies can sometimes use this
opportunity as timing option which creates an obvious potential for upward performance bias.
The practice of data vendors to backfill the return history of funds while adding a new fund to
their database creates the fourth form of data-conditioning bias, also known as an instant history bias.
A backfilling bias is closely related to self-selection bias, and sometimes these two biases are
integrated to each other (e.g., see Deaves, 2004). However, the distinguishing factor between them is
that a backfilling bias is caused by the practice of data vendors, whereas a self-selection bias stems
from omission of funds. Since underperforming funds are more prone to be excluded from databases
than are their outperforming counterparts, the sample of fund records to be backfilled biases average
initial performance upwards. Nevertheless, the influence of a backfilling bias on performance
persistence is not so clear since initial outperformance during the first recorded year may strengthen the
short-term persistence, but on the other hand, it may weaken the longer-term persistence. Therefore, a
backfilling bias might give a partial explanation why performance persistence is found more often
when relatively short selection and holding periods are employed in research design.
In addition, the research community is tempted to report results that are against market
efficiency than results supporting it (for excellent discussion of this tendency, see Black, 1993).
Therefore, it is presumable that the results of the studies published in financial journals are biased
towards showing performance persistence more often than found in all persistence studies made. It is
also clear that many more combinations of selection period and holding periods of various lengths may
have been tested than reported in journal articles (The bias of this kind stemming from the behavior of
scholars is known as data-snooping bias (e.g., see Lo and MacKinlay, 1990). Data-snooping is also
known as data-mining (e.g., see Black, 1993) or data-dredging (e.g., see Fama, 1991) who also
introduces the related concepts of model-dredging and factor-dredging which both might bias the
aggregate results of persistence studies as well). Thus, the direction of bias is most likely such that
results showing no persistence are omitted more often than those showing persistence.
When drawing conclusions from performance persistence studies it must be noted that the
results are always sample-specific and can not be generalized as such. First, based on the aggregate
results of the studies it is obvious that both the degree and direction of consistency in performance vary
over time. There are some time periods of clear evidence of persistence no matter what performance
metrics is employed. Correspondingly, there are other time periods for which almost all the
performance metrics show no evidence of persistence. On the contrary, results may indicate rather
18. International Research Journal of Finance and Economics - Issue 34 (2009) 134
performance reversal than persistence. The occasional mean-reversion effect documented in stock
returns (e.g., see DeBondt and Thaler, 1985, 1987; Fama and French, 1988; Poterba and Summers
1988; Malliaropulos and Priestley, 1999; Balvers et al., 2000; Chaudhuri and Wu, 2004; Balvers and
Wu, 2006; Ho and Sears, 2006; Nam et al., 2006) is also reported in several mutual fund studies (e.g.,
see Jain and Wu, 2000; Prather et al., 2004, for evidence from equity funds, and Jan and Hung, 2003;
Polwitoon and Tawatnuntachai, 2006, for evidence from bond funds). Both kind of consistencies
described above may arise from the market conditions that often favor some investment strategy over
some other until the conditions change. Due to the seasonality in persistence it is very difficult for a
mutual fund investor to find outperforming strategy on the basis of past performance.
Moreover, the evidence of persistence varies not only over time periods but also over markets during the
same time period. For example, Fletcher and Forbes (2002) find that much of the persistence of U.K. unit trust
performance is concentrated in 1980s, while as Malkiel’s (1995) results based on U.S. data show considerable
persistence during the 1970s but no persistence during the 1980s. Of course, the contrary results may also be
explained by differences in methodologies employed in detecting persistence. The existence of persistence
varies also across fund types; for example, for equity and bond funds, the aggregate results are quite diverse,
whereas for money markets, the results support quite unanimously performance persistence (e.g., see Dominian
and Reichenstein, 1997; Dahlquist et al. 2000; Jan and Hung, 2003) explained by small gross return differences
between the money market funds and the dominant role of expense ratio in determining the net return of money
market funds.
In addition, the optimal length of selection period on which the selection of fund or fund portfolio is
based seems to vary over time and it also seems to depend on not only the moment of decision-making, but also
on the methodology used in performance evaluation (e.g., see ter Horst and Verbeek, 2000; Jan and Chiu, 2007).
Though the general trend in the research design of the performance persistence studies has been towards shorter
selection and holding periods there is no unambiguous proof that shorter selection period would always increase
the prediction power of past performance (for the contrary proof, see Allen and Tan, 1999; Blake and Morey,
2000, for example). The most widely-used lengths of selection periods in the studies of the 2000s are one and
three years, but quite recently, also selection periods as short as 3 months have been adopted in the studies using
daily returns (e.g., see Bollen and Busse, 2005; Busse and Irvine, 2006).
The persistence literature seems to be quite unanimous that if performance persistence exists it
is rather short-term phenomenon ranging from one month (e.g., see Goetzmann and Ibbotson ,1994;
Harlow and Brown, 2006; Huij and Verbeek, 2007) to one year (e.g., see Hendricks et al., 1993;
Philpot et al., 2000; Droms and Walker, 2001a, 2001b; Jan and Hung, 2003, 2004; Polwitoon and
Tawatnuntachai, 2006) and in addition, that it can be to large extent explained by persistence in inferior
performance (e.g., see Hendricks et al., 1993; Shukla and Trzcinka, 1994; Blake and Morey (2000;
Quigley and Sinquefield, 2000; Detzler, 2002).
6. Summary
The lively debate on performance persistence of mutual funds continues among both scholars and
investment practitioners. The preceding review of persistence studies indicates that the direction of the
results often depends on the methodology and the performance model employed, as well as on the
sample data and the time period examined. Also the lengths of selection and holding periods affect the
results, and there is also inter-dependency between the period lengths and the methodology. The
general trend in the research design of the performance persistence studies has been towards shorter
selection and holding periods. This tendency coupled with the recent methodological refinements has
indisputably increased the proportion of the studies in which performance persistence is documented.
However, further evidence from longer time period is required to show that winning funds could be
identified ex ante by employing these advanced techniques in performance evaluation. On the other
hand, the shorter the holding period, the more difficult it is to economically benefit from performance
persistence due to increasing costs of more frequent rebalancing. In addition, there is hardly any
evidence that picking only the best-performing fund of the selection period would result in superior
performance in the subsequent holding period. At best, the odds to achieve better-than-average