method allows calculation of a one-year four-index alpha measure, and it provides the authors with a larger sample size,
which improves the chances of finding significant results.
Indeed, table 7b in Cremers et al. (2016) examines the relationship between active share and performance, using a sample of
346,711 fund-year observations.
In this paper, we take an alternative approach to Cremers et al.
(2016). Instead of using the Carhart (1997) method, we put ourselves in the position of a typical investor that buys and holds a
fund for a relatively long period of six years. Instead of pooling
annual samples, we choose all the emerging market equity
funds as of a certain time, i.e., December 31, 2008, and follow
them for the relatively long period of six years. 2 This approach
has the distinct disadvantage of using a smaller sample, but we
believe it has some clear advantages. First, the long time series
for each fund allows us to avoid the Carhart method of obtaining a one-year alpha. For the typical practitioner, Carhart
one-year alphas are hard to understand and almost never used.
Second, we do not have to pool the data because each fund has
a long time series of returns. We believe this is a distinct advantage because the pooling of the fund years requires dealing
with unbalanced pools that may have substantial serial correlation. Third, the long time series for each fund allows us to
examine the time series of active share for each fund. As a
result, we can now examine how the activeness of the fund
itself has changed over six years.
With the above as background, the specifics of our fund
selection process are as follows. To begin, we only include
diversified emerging market equity funds whose stated pro-
spectus benchmark index is the MSCI Emerging Markets
Index. We choose these funds for three reasons. First, they rep-
resent, by far, the largest number of emerging market equity
funds on the Morningstar database that shared the same bench-
mark. Second, existing size (back to January 2009), value, and
momentum indexes for the MSCI Emerging Markets Index
allow us to calculate a four-index alpha for these funds. These
factor indexes do not exist back to January 2009 for many of
the other emerging market equity benchmarks. Third, and most
importantly, as Frazzini et al. (2016) shows, active share in U.S.
equity funds is correlated with benchmark results. In other
words, Frazzini et al. (2016) show that funds with high active
share have benchmarks that consistently underperform the
benchmarks of funds with low active share. Therefore, funds
with high active share are outperforming because their bench-
mark returns perform poorly and funds with low active share
are underperforming because their benchmark returns perform
relatively well. Indeed, according to Frazzini et al. (2016), when
one examines funds within only one benchmark, active share
does not predict fund performance. In light of this, we choose
to examine only one style of fund (diversified emerging mar-
kets) with one prospectus benchmark (the MSCI Emerging
Markets Index). If we had chosen to include other funds we
would have had to use other benchmarks and our results may
have the problem described above, i.e., high active share funds
that are doing well because they are being compared to bench-
marks that underperform.
Using only funds that use the MSCI Emerging Markets Index as
the prospectus benchmark, we create our sample by taking
funds in existence on the Morningstar Principia Disk as of
December 31, 2008, with holdings data for the quarter ending
December 31, 2008 (so that we could calculate the active share
as of that date). We choose this time period because
Morningstar did not provide holdings data to calculate the
active share before the end of 2008. We eliminate any replicate
funds due to multiple share classes. This produced a sample of
sixty-seven funds. We follow these funds for the next six years
(until the end of 2014). For every quarter during the six years,
we calculate the active share of the fund using the holdings
data from Morningstar. If a fund’s quarterly active share data
disappears for more than one quarter3 or its monthly returns or
other fund characteristics (net assets, expense ratio, and turnover) discontinue due to a merger or liquidation, we consider
the fund to have dropped out of the sample. All but fourteen of
the sixty-seven funds survive the entire six-year period. For
each of the fourteen funds that drop out of the sample, we
assume that we have the fund’s actual returns and fund characteristics before it drops out. After the fund drops out, it takes on
the characteristics of the average surviving fund; i.e., after the
fund has dropped out, the monthly returns, active share, net
assets, expense ratio, and turnover of the fund are those of the
average surviving fund. This procedure ensures that our sample
includes funds that did not survive the six-year sample period.
Our main question in this paper is whether more-active funds,
as proxied by active share, produce better fund performance.
To answer this question, we regress the average active share
of the fund on fund performance over the sample period. This
is somewhat different from Cremers et al. (2016), which examines whether active share in one year predicts performance
in the following year. 4 We believe our methodology is an
improvement because we are examining active share contemporaneously with performance and as such can see more
clearly the relationship between active share and performance.
In the Cremers et al. (2016) method of using active share in
year t to predict fund performance in year t + 1, the active
share of the fund may have changed during the year when the
performance is measured.
Besides the average active share, we also use other variables
to control for size, expenses, turnover, and fund age (these
controls have been widely used in the mutual fund literature).
For the fund controls, we use the first available annual average
reported by Morningstar during the year. Fund age is age of the
fund, measured in years, as of January 1, 2009.