How to increase the odds

of owning the few stocks

that drive returns

Vanguard Research February 2019

Chris Tidmore, CFA; Francis M. Kinniry Jr., CFA; Giulio Renzi-Ricci; Edoardo Cilla

■ Some investment strategists advocate concentrated, “best ideas” portfolios as the surest
path to equity market outperformance. The premise is obvious: When a portfolio consists
only of a manager’s best ideas, returns are undiluted by second-best or lesser ideas. But
the reality is different.

■ We use simulations and empirical analysis to evaluate the relationship between portfolio
diversification and outperformance. Rather than raise the outperformance odds, increasing
concentration lowers them. The less diversified a portfolio, the less likely it is to hold
the small percentage of stocks that account for most of the market’s long-term return.
Concentration can increase the odds of earning high margins of outperformance, but the
probability of missing that return target increases more quickly than the probability of
reaching it.

■ Our analysis yields two measures that a manager must meet to outperform the market: the
“excess return hurdle” and the “success rate.” The excess return hurdle is the expected gap
between portfolio and market returns at different levels of concentration, and our analysis
shows this decreases with increased holdings. Success rate is a measure of the manager’s
ability to identify outperformers. The success rate necessary for a portfolio to outperform
decreases as the number of holdings increases.

Acknowledgment: The authors would like to thank Andrew S. Clarke, CFA, Daniel B. Berkowitz, CFA, and Lucas Baynes
for their contributions to this research.
 Many investors and financial professionals believe that excess returns are explained by the number of holdings.
broadly diversified portfolios are inferior to concentrated The empirical results are consistent with our theoretical
portfolios made up of a manager’s “best ideas.” This analysis, though they are notably time-period-dependent.
belief has been informed by research showing that
portfolio managers’ “top picks” have tended to
outperform the rest of their holdings (Cohen, Polk, and Discerning what an active manager’s
Silli, 2010, and Yeung et al., 2012). The premise seems best ideas were is not easy
straightforward: When assets are concentrated in a A simple review of an active manager’s portfolio can
manager’s best ideas, the performance of these yield misleading conclusions about the manager’s best
securities is undiluted by less promising ideas. Despite ideas. For example, it seems reasonable to assume that
the intuitive appeal of this premise, theoretical and the largest positions are the best ideas—those in which
empirical research fails to validate it. Why? the manager has the greatest conviction. A simple
illustration, though, shows why this assumption can be
We address this question in three steps. First, we misplaced. It argues against the conclusions drawn in
explore the concept that a portfolio’s best ideas can be research such as Yeung et al. (2012) that derived
extracted from a more diversified portfolio to create a concentrated results from diversified fund portfolios.
more concentrated one that will produce returns equal to
those of the “best ideas” subset. This assumption Consider an active manager that begins with five great
reflects hindsight bias that obscures the difficulty of ideas and allocates assets evenly to all five. At the end of
identifying a portfolio’s best ideas. the fund’s first year, Stock A has appreciated 30%, Stock
B 20%, and Stock C 10%, while Stock D has returned
Second, we conduct simulations to build randomly 0% and Stock E –10%. As shown in Figure 1, Stock A is
selected, equal-weighted portfolios with varying numbers now the largest holding and was the best performer for
of holdings. We find that the more broadly diversified the year, but it wasn’t the best idea; it was one of five
portfolios (those with relatively more holdings) outperform equally good ideas.
the more concentrated ones (those with relatively fewer
holdings but also a tighter dispersion of excess returns). The initial best ideas may also not start out intended as
Consistent with work by Ikenberry, Shockley, and such. A manager may be adept at adding to the more
Womack (1998); Heaton, Polson, and Witte (2017); and profitable positions over time. In this case, the “best idea”
Bessembinder (2018), our analysis highlights the risk of is not a permanent designation, but rather a dynamic status
excluding the minority of stocks that produce an outsized contingent on performance, new information, and/or
share of the market’s return in any given period. It also insight. Other managers may be good at cutting back or
yields parameters such as the “excess return hurdle” or eliminating their losing positions. Some best ideas may
the stock-selection “success rate” a manager must clear entail significant risk, and so the manager may proactively
to outperform at different levels of portfolio concentration. keep the position small. In each of these cases, it would be
inaccurate to conclude that the current largest positions
Finally, we complement the simulated analysis with an were the initial or current best ideas and should have been
empirical analysis of the historical performance of U.S. the sole investments in the manager’s portfolio.
mutual funds. The empirical analysis uses panel data
regression to estimate to what extent mutual fund

Notes on risk

All investing is subject to risk, including possible loss of principal. Past performance is no guarantee of future returns.
Diversification does not ensure a profit or protect against a loss. The performance of an index is not an exact
representation of any particular investment, as you cannot invest directly in an index.

2
 Figure 1. Larger positions may be due more to market Historical individual stock returns
performance than to initial best ideas We confirm the work by Ikenberry, Shockley, and
Womack (1998); Heaton, Polson, and Witte (2017);
and Bessembinder (2018), among others, which showed
0.25
that, because of the skewness of cumulative equity
0.20
returns, a minority of stocks are responsible for the
Portfolio weighting

market’s cumulative gains. The majority of stocks

0.15 throughout history have been relative losers. We
conducted a similar analysis by calculating the cumulative
0.10 returns of stocks in the Russell 3000 Index for January
1987 through December 2017, and the results were
0.05 consistent. Figure 2, on page 4, displays the frequency
of cumulative returns as well as the median and average
0.00 values, and it shows, like Cembalest (2014) and Edwards
Stock A Stock B Stock C Stock D Stock E
and Lazzara (2016), that approximately 47% of stocks
Initial allocation
End of year were unprofitable investments and that almost 30% lost
more than half their value. On the other hand, roughly
Note: The example shown is hypothetical and for illustrative purposes only; it is 7% of stocks had cumulative returns over 1,000%.
not based on any actual portfolio.
Source: Vanguard.
Figure 2 shows that the median stock’s lifetime return
was 7%, whereas the average stock in a portfolio
composed of all available stocks returned 387%. In fact,
as we start from one stock and add more stocks to the
Concentration, risk, and return:
portfolio, the portfolio return is more likely to improve
A bottom-up analysis
from the median to the average stock return. This is
Assessing the benefits of diversification within equities is because the probability of owning the market’s extreme
important, and significant research has been done on winners increases. This implies that by increasing a
how many stocks constitute proper diversification. portfolio’s number of randomly selected stocks, investors
Graham (1949), Evans and Archer (1968), Fisher and are more likely to include those stocks with higher
Lorie (1970), Malkiel (1973), and Statman (2004) mostly magnitudes of return. This is key, because the high-
addressed the effect of a portfolio’s size on its magnitude returns of a smaller number of stocks
nonsystematic risk but not on its return. Because the outweighs the lower—or even negative—returns of a
return distribution of stocks exhibits positive skew, we larger number of stocks.1
look to further address the question of concentration
from both risk and return perspectives using historical
stock performance.

1 In addition, long-horizon returns of stocks exhibit positive skew both because monthly returns are positively skewed and because the compounding of random returns
creates positive skew in a multiperiod distribution. For further details and a more in-depth discussion underlying a similar analysis and results, see Bessembinder (2018). 3
 Figure 2. The average stock’s cumulative lifetime A portfolio construction simulation
return significantly exceeds that of the median stock We build on the analysis presented above to create a
model for testing the performance of randomly selected
portfolios holding various numbers of stocks. The model
–100% to –50% is based on a theoretical framework in which we can
>-50% to 0% simulate a large number of portfolios that could have
Median 7% been created using historical stock returns.
>0% to 50%

>50% to 100% To start, we define concentration based on the number

of stocks in a portfolio, assuming that diversification
>100% to 150%
increases as that number rises. Therefore, we analyze
>150% to 200% randomly simulated portfolios of different sizes, from
one stock to 500 stocks.2
>200% to 250%

>250% to 300% After simulating the portfolios with different levels of

Stocks’ cumulative lifetime return

>300% to 350% diversification, our goal is to answer these questions: (1)

What is the probability that the portfolios will outperform
>350% to 400% Average 387% the benchmark? (2) What is the conditional average
>400% to 450% excess return of the portfolios that outperformed and
underperformed the benchmark? (3) What is the
>450% to 500% expected excess return of the portfolios? The answers
>500% to 550% can provide useful inputs to evaluate the risk and return
implications of holding a more or less diversified equity
>550% to 600%
portfolio. We also shift from addressing the probability
>600% to 650% of out- or underperforming a benchmark to having the
chance to outperform higher excess return targets across
>650% to 700%
changes in the number of holdings.
>700% to 750%
Our final objective is to study how portfolios’ risk and
>750% to 800%
return performance is affected across a range in the
>800% to 850% number of holdings when investors believe they have (or
have identified) stock-selection skill that leads to a higher
>850% to 900%
probability of picking top-performing stocks. We test the
>900% to 950% relationship between increased success rates (the
>950% to 1000%
probability of picking above-median performers) and
number of holdings and analyze their effect on the
>1000% likelihood and resulting magnitude of beating the
0 5 10 15 20 25 30 35
benchmark. This results in two useful tools for manager
evaluation: a manager’s excess return hurdle and the
Frequency
necessary success rate.

Note: Data cover January 1987 through December 2017.

Sources: Vanguard calculations, based on Russell 3000 Index constituents’ return
data from Thomson Reuters MarketQA.

2 Although holding few stocks (or, for example, one stock only) will lead to a concentrated portfolio, holding a larger number (such as 30 stocks) will not necessarily ensure
diversification. We understand that an investor might select all (or the majority of) the stocks from a specific market segment or, on the contrary, select each stock from
a different segment. We further understand there are other ways to measure the concentration level, including active share (Cremers and Petajisto, 2009), sector
4 concentration, factor concentration, and percent of holdings in the fund’s top ten holdings.
 Data and methodology To analyze the risk and return profile of the simulated
Our dataset encompasses the universe of stocks of portfolios, we make three assumptions:
the Russell 3000 Index from January 1987 through • We have equally weighted the stocks in every
December 2017.3 For each stock, we use monthly portfolio and have defined the benchmark as the
returns. Our analysis also assumes that a hypothetical equal-weighted market portfolio made up of all
investor would invest her or his capital over the entire stocks available at each point in time in the Russell
data sample period. 3000 Index.6 For the purpose of this analysis, and
thus to isolate the effect of concentration, it is
We create time series of portfolio performance as follows: important to use the same weighting method for
1. At the beginning of each quarterly rebalancing period, both portfolio and benchmark, to reduce potential
we randomly select stocks to form portfolios holding n bias resulting from stock characteristics (for example,
stocks where n equals 1, 5, 10, 15, 30, 50, 100, 200, small-capitalization bias).
or 500.4 Each stock must exist at the beginning of the • Should one or more stocks that have been selected
quarter and has an equal probability of selection. Each be removed from the benchmark during the investment
portfolio is created independently (so the same stock period, we assume that the corresponding capital is
could appear in more than one portfolio). invested in cash paying no interest.7
2. After selecting the stocks, we construct the portfolio • Because our goal is to build a theoretical framework
by equally weighting them. to compare levels of concentration in portfolios, our
3. We then compare each portfolio’s performance analysis does not consider management fees and
with that of the equal-weighted benchmark and transaction costs. Depending on their magnitude,
check whether the portfolio outperformed or including this information could lead to different results.
underperformed the benchmark. For each portfolio,
we also compute the excess return and tracking error Simulation results
versus the benchmark.5 We ran 10,000 simulations for each portfolio size to first
calculate the percentage of the portfolios that would
4. For each portfolio size, we conduct the above three
outperform the benchmark.
steps 10,000 times, and we compute average risk and
return summary statistics.

3 We understand that using the stocks in a segment of the Russell 3000 Index, such as the Russell 1000 Index or Russell 2000 Index, could lead to different results.
4 We also tested annual rebalancing with similar results.
5 We define tracking error as the annualized standard deviation of excess returns.
6 We believe that using the stocks included in the Russell 3000 Index universe can help avoid the problem of picking microcapitalization stocks that cannot be easily
traded.
7 We believe this assumption is reasonable, as reallocating the capital to the remaining stocks could increase the portfolio’s concentration during the rebalancing period. 5
 Figure 3 shows that the probability of outperforming the Figure 3. The simulated portfolios’ probability
benchmark rises rapidly from one-stock portfolios to ones of outperforming the benchmark rises
with more holdings before leveling off to a gradually with an increase in holdings
increasing probability of outperformance.8 For instance,
that probability rises from 11.1% for one-stock portfolios
50 48.4%
to 48.4% for 500-stock portfolios.9 All else being equal, 44.7%45.6%

Probability of outperformance
42.7%
our results suggest that holding a higher number of 40.3%
40 37.3%
stocks increases the chances of outperforming. 34.5%
28.7%
30
Up to now, we have studied only the probability of
outperformance, without providing information on the 20
magnitude of out- or underperformance. Figure 4
11.1%
displays the average excess return conditional both on 10
out- and underperformance for each of the portfolios.10
As an example, for the 30-stock portfolio simulations, 0
1 5 10 15 30 50 100 200 500
the average return is 1.1% for portfolios that outperformed
Number of stocks in the portfolio
and –1.4% for portfolios that underperformed.
Note: Data cover January 1987 through December 2017.
The results show that portfolios with fewer stocks Sources: Vanguard calculations, based on quarterly Russell 3000 Index
tend to be characterized by both higher performance constituents’ return data from Thomson Reuters MarketQA.
dispersion and a pronounced negative asymmetry
in returns. As the number of stocks increases, the
dispersion narrows, and the negative asymmetry shrinks.
Figure 4. The dispersion of the average conditional
Average expected excess return: excess return shrinks with an increase in holdings
A portfolio’s “excess return hurdle”
By combining information from Figure 3 on portfolios’
probability of outperforming the benchmark and from 6%
Average conditional excess return

4.2%
Figure 4 on conditional average performance, we can 2.4% 1.9%
3 1.5% 1.1% 0.9%
compute the expected excess return of the simulated 0.6% 0.5% 0.3%
portfolios in Figure 5. For example, the average excess 0
–0.3%
–1.0% –0.7% –0.5%
return for the 30-stock outperforming portfolios is 1.1% –3 –2.0% –1.4%
–2.6%
(Figure 4), and we multiply that by the outperformance –6
–3.8%
probability of 40.3% (Figure 3). We then do the same for
–9
the 30-stock underperforming portfolios, multiplying
–1.4% by 59.7%. Finally, we sum the two results to get –12
–11.7%
the average annualized conditional excess return of –15
–0.4% for a 30-stock portfolio as seen in Figure 5. These 1 5 10 15 30 50 100 200 500
values can be treated as the excess return hurdle that a Number of stocks in the portfolio
portfolio must clear to match the benchmark return. In Outperforming portfolios
other words, our simulations suggest that a portfolio of Underperforming portfolios
30 randomly selected stocks can be expected on average
Notes: Data cover January 1987 through December 2017. Portfolios are rebalanced
to trail the benchmark by 0.4%. A portfolio manager
quarterly, and annualized average returns are presented separately for
must have at least enough skill to overcome this cost outperforming and underperforming portfolios for each portfolio size.
of concentration. Sources: Vanguard calculations, based on quarterly Russell 3000 Index
constituents’ return data from Thomson Reuters MarketQA.

8 We have also tested portfolios with a higher number of holdings, up to the extreme case of n–1 stocks, where n is the number of stocks in the benchmark at any point
in time. In this case, the probability of outperforming the benchmark converges to 50%; this is justified by the fact that we defined the benchmark as the arithmetic
average of all stocks in the universe.
0.8 0.8
9 When we ran the simulations over longer periods, the probability of outperformance decreased across all numbers of stock holdings, and the fewer the holdings, the
greater the magnitude of decline. 0.7 0.7
6 10 We also analyzed median performance with quantitatively similar results. 0.6
0.6
0.5 0.5
 Figure 5. The average expected excess return of the More concentration, greater tracking error
simulated portfolios rises with an increase in holdings In addition to return-based measures, our simulations
estimated tracking errors. Figure 6 displays the average
tracking error for each portfolio size. For each of the
0%
Average expected excess return

–0.2% –0.1% –0.1% 0.0% concentrated portfolios, the relationship is inversely

–0.7% –0.4%
–1.0%
–2
–2.0%
monotonic: Increasing the number of holdings decreases
the average tracking error.
–4

–6 We find that higher numbers of randomly selected

portfolio holdings are associated with increased chances
–8
of outperforming the benchmark, higher average excess
–10 returns, less-negative excess returns, and lower tracking
–9.9%
error. These findings make a compelling case for broadly
–12
1 5 10 15 30 50 100 200 500
diversified active portfolios. In this sense, we agree with
Number of stocks in the portfolio
Edwards and Lazzara (2016), which focused on the risks
of active managers following more concentrated
Note: Data cover January 1987 through December 2017. strategies. Lastly, although it is important to understand
Sources: Vanguard calculations, based on quarterly Russell 3000 Index that more diversification alone does not guarantee better
constituents’ return data from Thomson Reuters MarketQA. results, diversification when paired with stock-selection
skill can enhance the odds of outperformance.

As expected, given the results shown in Figures 3 and 4,

the average excess return is significantly negative for
Figure 6. The average tracking error of the
the one-stock case and becomes less negative when
simulated portfolios declines as the number
we increase the portfolios’ number of stocks, till it
of holdings increases
becomes very close to the benchmark with the 500-
stock portfolios. As Figure 5 shows, expected average
excess return goes from an annualized –9.9% for the 50%
44.4%
one-stock portfolios to 0.0% for the 500-stock portfolios.
Average tracking error

40
These results tie back to Figure 2, and prior research
by Ikenberry, Shockley, and Womack (1998); Heaton, 30

Polson, and Witte (2017); and Bessembinder (2018) 20.0%

20
shows that picking one random stock (equivalent to 14.2%
11.6%
picking the median stock) will most likely mean 8.2%
10 6.3%
underperforming the average of all stocks, while 4.4% 3.1%
1.8%
choosing more stocks will move an investor toward 0
the average. 1 5 10 15 30 50 100 200 500
Number of stocks in the portfolio

Note: Data cover January 1987 through December 2017.

Sources: Vanguard calculations, based on quarterly Russell 3000 Index
constituents’ return data from Thomson Reuters MarketQA.

7
 Probability of achieving higher excess return targets levels increases significantly as we reduce the number of
We have looked at the probability of generating positive holdings. Figures 7a and 7b show that a portfolio with
or negative excess returns. We understand that some 100 holdings has a 10% chance of outperforming by
investors’ objective may be not just to beat the greater than 1% annually and a 14% chance of
benchmark, but also to increase the probability of underperforming by 1%. Decreasing the holdings to 30
excess returns above a certain target. Theoretically, increases the probability of outperforming by 1% or
this should increase with greater concentration. more by 9 percentage points, to 19%, but it increases
What we find, however, is that this higher chance of the probability of underperforming by 1% or more by 20
outperformance comes with an even higher probability percentage points, to 34%.
of underperformance by the same excess return target.
Investors who seek potential higher return and are
Using our random selection simulation results, we can see comfortable with the resulting higher risk on their equity
in Figure 7a that as the number of holdings starts to allocation may choose a more concentrated manager
decrease, the likelihood of outperforming return targets over a more diversified one. In contrast, investors who
slowly rises, but the increased probability of outperforming are more sensitive to risk—particularly active risk—may
levels off before decreasing significantly. This contrasts wish to allocate to more diversified managers.
starkly with the results in Figure 7b, which shows that the
probability of underperforming certain negative return

Figure 7. The relationship between holdings and the probability of achieving higher return targets
is nonlinear

a. The probability of outperforming return targets above the benchmark increases slowly before declining as
holdings decrease

Number of holdings
Return target
above the
benchmark 1 5 10 15 30 50 100 200 500
5% 4% 4% 2% 1% 1% 0% 0% 0% 0%
4% 5% 6% 4% 2% 1% 1% 0% 0% 0%
3% 6% 9% 8% 5% 2% 1% 1% 0% 0%
2% 8% 14% 14% 11% 7% 4% 1% 1% 0%
1% 9% 21% 23% 22% 19% 16% 10% 4% 1%

Worse to better outcomes

b. The probability of underperforming return targets below the benchmark increases as holdings decrease

Number of holdings
Return target
below the
benchmark 1 5 10 15 30 50 100 200 500
–1% 86% 60% 51% 44% 34% 25% 14% 5% 0%
–2% 84% 50% 36% 27% 14% 6% 1% 0% 0%
–3% 81% 39% 27% 14% 3% 0% 0% 0% 0%
–4% 77% 29% 13% 6% 0% 0% 0% 0% 0%
–5% 73% 20% 6% 2% 0% 0% 0% 0% 0%

Worse to better outcomes

Notes: Data cover January 1987 through December 2017.

Sources: Vanguard calculations, based on quarterly Russell 3000 Index constituents’ return data from Thomson Reuters MarketQA.

8
 Skill, luck, and probability theory Figure 8. The per-stock hit rate required for a
Whether the majority of the performance of active 90% chance of portfolio outperformance increases
managers is driven by skill or luck is a question beyond as the number of stocks decreases
our research scope.11 Skilled managers are out there,
but nobody is close to perfect. A manager who has a 100%
persistent edge would like to have more opportunities, 90%

not fewer, to showcase it.

Per-stock success rate

80 74%
63% 60%
One straightforward example is that of a tennis player, 60 57% 55% 53%
Jill, who is slightly better than another player, John, and
might win 53% of any single points played. Jill would 40
like to play as many points as possible to take advantage
20
of her slight edge in tennis skill. If John and Jill played a
three-set match, Jill might win it 76.8% of the time.12
0
Think of the points played as the number of investments. 1 10 30 50 100 200 500
Number of stock picks
Probability theory says that if you have a small edge, that
edge can be amplified by making more bets, not fewer. Note: The example shown is hypothetical and for illustrative purposes only.
An above-average manager (one with a “success rate,” Source: Vanguard.
or probability of picking an outperforming stock, above
50%) should make as many bets as possible (Grinold,
1989). This is not dissimilar from Jill the tennis player. An
active manager who has a 53% success rate (probability
benefits of a persistent edge. The more diversified
of winning a point) and picks 500 stocks has a 90%
manager needs a lower hit rate, or success rate, to
likelihood that at least 251—a majority—of those stock
perform the same as a more concentrated manager with
picks will outperform. Another manager who makes only
a higher rate. This conclusion does get cloudy when the
ten picks would need a 74% success rate to achieve the
hit rate diminishes as the number of bets gets larger,
same 90% likelihood that a majority of those picks will
which we might find with a fund of limited resources.
outperform. Figure 8 shows that as the number of stock
The question remains, how does a better batting average
picks decreases, the hit rate needed to achieve the
manifest itself in manager excess returns? Sorensen,
same probability of success increases.
Miller, and Samak (1998) estimated that a quarterly
batting average of 52% equated to an excess-return
Probability theory highlights flaws in the common claim
range of 1.40% to 3.02% and that a quarterly 54%
that diversified strategies lack conviction and are
batting average equated to excess returns of 2.61%
benchmark-huggers. These strategies can more
to 5.59%.
appropriately be described as an effort to maximize the

11 For further insight into skill and luck, see Mauboussin (2012).
12 This assumes that the average number of points played is 60 per set, or 180 per three-set match. We are also assuming that winning a majority of the points would
constitute a win. Therefore, we calculate the cumulative binomial probability distribution in which the number of trials equals 180, the number of events is set to 91,
and the event probability is 53%. 9
 Investor success: Improving the success rate This approach echoes Sorensen, Miller, and Samak
Our research also aims to study how the probability—and (1998), whose research randomly simulated portfolios
magnitude—of outperforming the benchmark changes made up of 100 stocks. Their main purpose was to
across concentration levels when an investor manages to analyze the level of outperformance resulting from
select top-performing stocks. For this analysis, we assume various degrees of active manager success in selecting
that a top-performing stock is one whose excess return top-performing stocks. Similarly, our research looks at
over the rebalancing period is higher than that of the changing the success rates, and computing return and
median stock.13 In this case, we assume that an investor risk measures of the simulated portfolios and comparing
can, for example, consistently select a top-performing them with the base case, which is a manager success
stock with a probability equal to 52% at every rebalancing equal to 50% (that is, random stock selection). Our
period (quarterly in our case) for three decades. This framework expands on this to look at changing success
implies no decrease in the probability as we increase the rates across various portfolio sizes.
number of holdings, which may be more realistic for some
strategies and managers and less for others.

Figure 9. The probability that the simulated portfolios outperform the Russell 3000 Index goes up as success rate
and portfolio size increase

100%
Probability of outperformance

80

60

40

20

0
1 5 10 15 30 50 100
Number of stocks in the portfolio
Success rate
50%
50.5%
51%
51.5%
52%

Note: Data cover January 1987 through December 2017.

Sources: Vanguard calculations, based on quarterly Russell 3000 Index constituents’ return data from Thomson Reuters MarketQA.

13 This is just one potential definition of success. Different definitions (for example, the probability of selecting any of the top 10% performing stocks, or of avoiding the
worst 20% performing stocks) would lead to different results. It is also important to stress that the stocks in the top or bottom 50% set can be selected with equal
probability. This means an investor with a higher success rate will be able to pick a top-performing stock with, for example, a 52% probability; however, any of the top-
10 performing stocks will have the same chance of being selected.
 To understand our results’ sensitivity to the success rate, outperforming. In a 100-stock portfolio, by contrast, this
we chose five success rates at increments of 0.5%, same success rate yields an almost 100% probability of
starting from the base-case success rate up to 52%. outperforming.
Assuming a success rate equal to 50% implies that both
top- and bottom-performing stocks can be picked with In Figure 10, we report the expected excess return of
the same probability. Therefore, our results will be very portfolios simulated with various success rates. These
close to those shown in the previous sections.14 results help us answer the following question: What is
the success-rate level needed to achieve a specific
Figure 9 shows that both higher success rates and excess return target for a given portfolio size? For a five-
greater diversification are associated with a higher stock portfolio, a success rate as high as 52% is needed
probability of outperformance. In our simulation, to have a positive expected performance at the end of
however, diversification seems to have a more powerful the entire investment horizon. As we add more stocks to
impact. Even at a 52% success rate, the five-stock the portfolio, the success rate needed to outperform
portfolio barely has a greater than 50% chance of

Figure 10. The average expected excess return rises as success rate and portfolio size increase

4%
Average expected excess return

-2

-4

-6

-8

-10
1 5 10 15 30 50 100
Number of stocks in the portfolio
Success rate
50%
50.5%
51%
51.5%
52%

Note: Data cover January 1987 through December 2017.

Sources: Vanguard calculations, based on quarterly Russell 3000 Index constituents’ return data from Thomson Reuters MarketQA.

14 The difference in results is explained by random sampling approximation. 11

 decreases. In fact, for portfolios with 30 or more stocks, Data and methodology
any success rate higher than 50% considered in our We collected active U.S. equity funds’ quarterly data
analysis would lead to an excess return higher than zero. from January 2000 to December 2017 from Morningstar,
Inc. We selected only funds available for sale in the
Overall, the results show that holding a more diversified United States. Similar to Kacperczyk, Sialm, and Zheng
portfolio helps increase the chances of outperforming the (2005) and Goldman, Sun, and Zhou (2016), we also
benchmark and achieving higher average returns. limited our dataset to one share class per fund to
Investors who seek to outperform above certain excess make sure that multiple-share-class funds were not
return thresholds will improve their chances by reducing overweighted in our analysis. Lastly, funds-of-funds and
the number of holdings but will experience a more rapid funds with fewer than ten holdings were excluded.15
increase in the chance of underperforming. Lastly, the Ultimately, after adjustments, our final dataset includes
benefits of diversification are present and potentially 2,136 funds.
amplified in the case of an investor who can consistently
pick top-performing stocks. We define number of stock holdings as the number that a
fund held at the beginning of each measurement period
and analyze the impact of that number on fund excess
Historical approach to analyzing
returns. For the panel data analysis, we specify an
the concentration within portfolios
unbalanced time-fixed effects model in which we regress
So far we have evaluated concentration based on the funds’ excess returns using the Russell 3000 Index
simulated analysis using historical equity return data. as the relevant benchmark in each quarter on the number
In this section, we investigate, through historical fund of holdings, the Sector Concentration Index (SCI), the
performance, whether a relationship exists between active Factor Concentration Index (FCI), and a set of control
equity funds’ excess returns and their concentration. variables.16,17 We believe the introduction of “time
We conduct panel data regression to see how much of effects” allows us to better account for shared factors
excess return variance is explained by number of holdings and business-cycle events that might affect all funds.
while controlling for other variables, including sector and This ultimately leads us to define the following variables:
factor concentration.
• Net ExcRet is the net excess return of any fund.
These controls are important because having a large • Ln(Age) is the natural log of fund age.
number of stocks in a portfolio does not necessarily
imply low concentration. A significant fraction of the • Expense is the fund expense ratio.
holdings might come from the same industry sector • Turnover is annualized fund turnover.
(such as financials or technology) or have similar factor
exposure (such as value or small-cap bias). The results • Ln(TNA) is the natural log of fund total net assets.
yield two notable insights. First, there is a positive and • Ln(NumHold) is the natural log of number of fund
highly significant relationship between the number of stock holdings.
holdings and net excess returns. The more holdings, on
average, the higher the excess returns. Second, there is • SCI is the fund sector index as specified in
a negative and highly significant relationship between the Appendix I, on page 17.
number of holdings and fund expense ratios. On average, • FCI is the factor concentration index as specified in
more concentrated funds cost more. We now review the Appendix I, on page 17.
data and methodology in detail.

15 Although we tested funds with fewer than ten holdings in our bottom-up simulations, either they were funds-of-funds or the holdings were not individual stocks.
16 We ran the Durbin-Wu-Hausman test to verify the potential benefit of using a random effects model. (This test assesses the consistency of an estimator when
compared with a less efficient consistent estimator [Greene, 1997]; the test is often used for panel data analysis to choose between fixed effects and random effects
models.) Our findings fail to reject the null hypothesis, suggesting that a time–random effects model should be preferred. However, the results following this approach
are not quantitatively different from those following our time–fixed effects model.
17 We use the Russell 3000 Index, rather than the primary prospectus benchmark, as the equity benchmark to allow for consistency among funds. Also, to perform the
sector concentration and factor concentration data analysis, we need a unique and homogeneous market benchmark. We consider the Russell 3000 Index to be an
12 appropriate and widely accepted benchmark for U.S. equities.
 Since most of our variables show some level of positive Summary statistics and results
skewness with a few of the variables significantly Figure 11a, on page 14, documents the summary
affected by extreme values, we take the natural logarithm statistics for natural log number of holdings, sector
of the funds’ age, total net assets, and number of concentration, factor concentration, and other fund
holdings in order to linearize their relationship with excess characteristics. The mutual funds in our sample vary
return.18 To alleviate the potential impact of endogeneity, widely in their characteristics (for the non-log fund
we lagged all independent variables by one quarter. Also, characteristics, see Appendix II on page 18). Figure 11b
for robustness, we estimate our model using Restricted shows the contemporaneous correlations between the
Maximum Likelihood Estimation (REML):19 independent variables used in our model. On average,
the correlation between age and TNA is positive, as we
Net ExcReti,t = α + β1Ln(Age)i,t-1 + β2Expensei,t-1 + might expect given that older funds tend to have
β3Turnoveri,t-1 + β4Ln(TNA)i,t-1 + β5Ln(NumHold)i,t-1 + acquired more assets than younger funds. We also note
β6SCIi,t-1 + β7 FCIi,t-1+μt-1 + εi,t-1 that expense ratio has a positive correlation with factor
and sector concentration but a negative correlation with
Although our methodology is based on literature such as number of holdings. Overall, our dataset summary
Kacperczyk, Sialm, and Zheng (2005); Brands, Brown, statistics are consistent with previous studies (such as
and Gallagher (2005); Huij and Derwall (2011); and Kacperczyk, Sialm, and Zheng, 2005, and Goldman, Sun,
Goldman, Sun, and Zhou (2016), our approach differs in and Zhou, 2016).
three main aspects:

• We explicitly take into account in the panel model the Figure 12, on page 15, summarizes our findings for
number of holdings. different models’ specifications with and without
accounting for the effects of sector and factor
• We control for factor and sector concentration using concentration.20 The first point to notice is that most of
consistent definitions, and we include factor our estimates are statistically significant at a 1% level.21
concentration as a control variable rather than using The R-squared figures for our model specifications might
factor-adjusted funds’ alpha as the dependent variable. appear weak at first glance, but the amount of
• Our data sample is significantly larger than that of unexplained variance is not surprising. The low R-squared
previous studies, making our results more statistically is to be expected for active equity funds and does not
accurate. invalidate the significance of the explanatory variables.

18 Prior research is mixed on whether to take the natural logarithm of portfolio turnover when portfolio turnover is being used as a control variable. We tested our model
using both methods with similar results.
19 For further details, see Greene (1997).
20 We addressed other potential model specifications, including the removal of age or TNA or both, with quantitatively similar results.
21 The number of observations is considerably larger (over 100,000) than previous studies because of a combination of factors—namely, the number of funds, the time
period, and the data frequency (such as quarterly observations vs. annual observations). For instance, Kacperczyk, Sialm, and Zheng (2005) used a dataset with roughly
35,000 observations depending on the test and regressions performed. Brands, Brown, and Gallagher’s (2005) sample size was approximately 1,200 observations, and
Goldman, Sun, and Zhou’s (2016) dataset considered 34,176 observations. This has the advantage of making our coefficient estimates more reliable but leads us to
easily reject the null hypothesis for relationships that are not necessarily economically significant (Lin, Lucas, and Shmueli, 2013). 13
 Figure 11. Historical analysis summary statistics, January 2000–December 2017

a. Fund characteristics

Variable Mean Median 5th percentile 95th percentile

Total number of funds 2,136
Ln(Age) (years) 2.48 2.55 1.07 3.74
Expense (%) 1.19 1.14 0.62 1.95
Turnover (%) 86.89 61.00 12.00 214.00
Ln(TNA) (USD millions) 5.18 5.22 1.83 8.32
Ln(NumHold) (units) 4.40 4.32 3.40 5.80
SCI (%) 28.56 23.78 8.48 78.52
FCI (%) 42.51 40.45 13.46 73.51
ExcRet (%) 0.38 0.19 (5.89) 7.33

b. Correlation structure

Variable Ln (Age) Expense Turnover Ln(TNA) Ln(NumHold) SCI FCI

Ln(Age) 1.00
Expense -0.14** 1.00
Turnover -0.12** 0.09** 1.00
Ln(TNA) 0.47** -0.38** -0.17** 1.00
Ln(NumHold) 0.03** -0.23** 0.00 0.20** 1.00
SCI -0.01** 0.21** 0.21** -0.10** -0.30** 1.00
FCI -0.15** 0.22** 0.02** -0.11** 0.06** 0.13** 1.00
**Significant at 1%.
Notes: We used quarterly data for the oldest share class of U.S. active equity funds available for sale in the United States with more than ten holdings and excluding funds-
of-funds. Ln(Age) is the natural log of the number of years since the fund began. Expense ratio (Expense) is the annualized difference between quarterly gross and net excess
returns. Observations for which Expense is negative are omitted from the dataset. Turnover ratio (Turnover) measures the fund’s trading activity and is reported annually by
Morningstar, Inc.; quarterly observations have been computed by linearly interpolating annual observations. Ln(TNA) is the natural log of a fund’s total asset base, net of fees
and expenses. Observations for which TNA are less than $500,000 are omitted from the dataset. Ln(NumHold) represents the natural log of the number of stocks held by the
fund. Sector concentration (SCI) is annualized and represents the level of concentration in Global Industry Classification Standard (GICS) sectors. Factor concentration (FCI) is
annualized and represents the level of concentration in 9-box Morningstar style factors. Excess return (ExcRet) is the quarterly return in excess of the Russell 3000 Index,
before fees. Our data sample presents 19 observations in which the SCI is higher than 100% because of some funds applying leverage. This represents a minor proportion
(0.02%) of the total sample size (98,790).
Sources: Vanguard calculations, using data from Morningstar, Inc.

14
 Figure 12. Panel data regression: Time–fixed effects on net excess returns

Coefficient NumHold NumHold + SCI NumHold + FCI NumHold + SCI + FCI

Ln(NumHold) 9.49** 16.06** 4.24** 10.32**

Control Variables

Ln(Age) 4.60* 2.92 8.85** 8.38**

Expense –50.94** –74.27** –116.85** –133.94**
Turnover –0.05** –0.08** -0.05** –0.07**
Ln(TNA) –8.88** –8.74** –8.83** –9.13**
SCI 1.01** 0.82**
FCI 1.45** 1.36**
Adjusted R2 8.46% 8.77% 9.01% 9.11%

Notes: We used quarterly data for the oldest share class of U.S. active equity funds available for sale in the United States with more than ten holdings and excluding funds-
of-funds. Observations for which Expense is negative or TNA is less than $500,000 are omitted from the data sample. All regressions include time dummies. One star indicates
significance at the 5% level; two stars indicate significance at the 1% level. Fund age (Age) is in years; expense ratio (Expense) is in percentage points; turnover ratio
(Turnover), sector concentration (SCI), and factor concentration (FCI) are in percentage points; total net assets (TNA) is in millions of U.S. dollars; and number of holdings
(NumHold) is in units.
Sources: Vanguard calculations, using data from Morningstar, Inc.

The coefficient of number of holdings is statistically the other control variables (turnover, fund age, and total
significant at least at 5% across all model specifications net assets) are somewhat mixed relative to findings in
and is economically significant and positive, suggesting previous studies.22
that given any specified level of sector or factor
concentration, a fund’s performance improves with a Lastly, the time–fixed effects model specified above
higher number of holdings. This finding is consistent with allows for a change in the constant (α) across time
Goldman, Sun, and Zhou (2016). For example, based on but does not map any dynamic change in the beta-
Figure 12, increasing holdings by 10 percent leads to an coefficients. Additional analysis (see Appendix III on page
increase of 3.80 basis points (bps) in annual excess 19) does show that the effect of concentration (measured
returns (9.49 x 10% x 4). When we consider the impact by number of holdings) on excess return has fluctuated
of both sector and factor exposures in the fourth column considerably over time. This is consistent with Hickey et
of Figure 12, using the same 10% increase in holdings, a al. (2017), who found that U.S. large-cap equity funds
slightly higher coefficient of 10.32 leads to an increase of with fewer holdings had historically slightly higher returns
4.13 bps in annual excess returns. but that this was largely driven by short time windows in
the late 1990s and early 2000s, and we found similar
Consistent with Rowley, Harbron, and Tufano (2017), on time-period-specific outperformance by concentrated
average, expense ratio has a statistically significant funds. The overall impact of number of holdings on U.S.
negative impact on performance, and the coefficients of equity active funds’ excess returns is positive but
episodically has turned negative.

22 The coefficient of turnover is consistent with Kacperczyk, Sialm, and Zheng (2005); Goldman, Sun, and Zhou (2016); and Brands, Brown, and Gallagher (2005), while
fund age and total net assets are consistent with Brands, Brown, and Gallagher. 15
 Implications for investors and further research target, we found that decreasing the number of holdings
As with all investments, one should expect premiums if increased the chance of outperformance but came with an
taking on more risk. Common examples include equity vs. even higher probability of underperformance by the same
fixed income, corporate bonds vs. government bonds, and excess return target. In addition, investors who believe
high-yield bonds vs. investment-grade bonds. We found their stock-selection ability is better than chance would
that less diversified portfolios have more relative risk than be best served applying that skill by selecting more
more diversified ones and that investors should therefore stocks, not fewer. Finally, we tested mutual fund
expect higher returns from the less diversified portfolios, performance as a function of various levels of portfolio
but the evidence from our bottom-up simulations shows concentration as measured by number of holdings and
that concentrated portfolios of randomly selected stocks found that, historically, increased diversification yielded
have lower average returns than diversified portfolios. Our higher returns.
historical analysis is consistent with these theoretical
results, finding lower average returns for the less References
diversified portfolios. Our results suggest that a
concentrated manager creates more opportunity for Bessembinder, Hendrik, 2018. Do Stocks Outperform Treasury
Bills? Journal of Financial Economics (Forthcoming).
outsized excess returns (either positive or negative),
although as we decrease the number of holdings, the Brands, Simone, Stephen J. Brown, and David R. Gallagher,
chance of underperformance increases at a faster rate 2005. Portfolio Concentration and Investment Manager
than the chance of outperforming. Performance. International Review of Finance 5(3-4): 149–174.

Cembalest, Michael, 2014. The Agony and the Ecstasy: The

The mere function of being less diversified via random
Risks and Rewards of a Concentrated Stock Position. Eye on the
selection makes for a performance hurdle that portfolio
Market: Special Edition. J.P. Morgan Asset Management.
managers must overcome.23 So one must find talented
active managers with the ability to overcome this hurdle Cohen, Randolph B., Christopher Polk, and Bernhard Silli, 2010.
(along with typically higher fees). Those managers who Best Ideas. Working Paper. London: Paul Woolley Centre, The
can select a concentrated portfolio of “home run” stocks London School of Economics.
have the potential to earn extreme positive returns. The
Cremers, K.J. Martijn, and Antti Petajisto, 2009. How Active Is
question is whether they can reliably and consistently
Your Fund Manager? A New Measure That Predicts
identify those stocks in advance. This analysis yields Performance. The Review of Financial Studies 22(9): 3329–3365.
valuable tools for manager evaluation: excess return
hurdles or success-rate hurdles that can be used in Edwards, Tim, and Craig J. Lazzara, 2016. Fooled by Conviction.
conjunction with well-established determinants of Index Investment Strategy, S&P Dow Jones Indices.
success such as cost.
Evans, John L., and Stephen H. Archer, 1968. Diversification and
the Reduction of Dispersion: An Empirical Analysis. The Journal
of Finance 23(5): 761–767.
Conclusion
Historical cumulative returns of individual stocks are Fisher, Lawrence, and James H. Lorie, 1970. Some Studies of
skewed whereby overall market returns are determined Variability of Returns on Investments in Common Stocks. The
Journal of Business 43(2): 99–134.
by a small minority of stocks. Therefore, all else being
equal, a more diversified portfolio is more likely to hold Goldman, Eitan, Zhenzhen Sun, and Xiyu (Thomas) Zhou, 2016.
these outperforming stocks while displaying a lower The Effect of Management Design on the Portfolio
dispersion of portfolio returns. We conducted simulations Concentration and Performance of Mutual Funds. Financial
of various portfolio sizes and showed that those Analysts Journal 72(4): 49–61.
portfolios with fewer holdings underperformed those
Graham, Benjamin, 1949. The Intelligent Investor, first edition.
with more holdings, leading to a higher return hurdle to
New York, N.Y.: Harper & Brothers.
overcome. Understanding that some investors may
prefer to generate returns above a certain excess return

16 23 A full assessment of the optimal sizing of portfolios based on investors’ goals and preferences should be explored in further research.
Greene, William H., 1997. Econometric Analysis, third edition. Appendix I. Defining sector and factor
Upper Saddle River, N.J.: Prentice-Hall. concentration
Grinold, Richard C., 1989. The Fundamental Law of Active Similar to what was previously done by Brands, Brown,
Management. The Journal of Portfolio Management 15(3): and Gallagher (2005), Kacperczyk, Sialm, and Zhang
30–37. (2005), and Goldman, Sun, and Zhou (2016), we define
the Sector Concentration Index (SCI) at any point in time
Heaton, J.B., N.G. Polson, and J.H. Witte, 2017. Why Indexing
for any fund (i) as:
Works. Applied Stochastic Models in Business and Industry
33(6): 690–693. 11
SCIi = 1/2 ∑ |wi,GICS – wR3000,GICS |
Hickey, Michael, Christopher Luongo, Darby Nielson, and GICS=1
Zhitong Zhang, 2017. Does the Number of Stocks in a Portfolio A fund’s sector concentration is therefore the sector
Influence Returns? Fidelity Investments. active share (Cremers and Petajisto, 2009) compared
Huij, Joop, and Jeroen Derwall, 2011. Global Equity Fund with the relative sector weight of the Russell 3000 Index,
Performance, Portfolio Concentration, and the Fundamental Law which we use as the benchmark. For our analysis, we
of Active Management. Journal of Banking & Finance 35(1): use the 11 GICS industry sectors as of December 31,
155–165. 2017: consumer discretionary, consumer staples, energy,
financials, health care, industrials, information technology,
Ikenberry, David L., Richard L. Shockley, and Kent L. Womack,
materials, real estate, telecommunication services, and
1998. Why Active Fund Managers Often Underperform the S&P
utilities.
500: The Impact of Size and Skewness. The Journal of Private
Portfolio Management 1: 13–26.
We follow a similar approach for the Factor
Kacperczyk, Marcin, Clemens Sialm, and Lu Zheng, 2005. On Concentration Index (FCI):
the Industry Concentration of Actively Managed Equity Mutual 9
Funds. The Journal of Finance 60(4): 1983–2011. FCIi = 1/2 ∑ |wi,factor – wR3000,factor |
factor=1
Lin, Mingfeng, Henry C. Lucas Jr., and Galit Shmueli, 2013. Too
Big to Fail: Large Samples and the p-Value Problem. Information We define a fund’s factor concentration as the 9-box
Systems Research 24(4): 906–917. Morningstar style factor active share compared with the
relative weight of the Russell 3000 Index. These are:
Malkiel, Burton, 1973. A Random Walk Down Wall Street, first small-cap value, mid-cap value, large-cap value, small-cap
edition. New York, N.Y.: W.W. Norton & Company. blend, mid-cap blend, large-cap blend, small-cap growth,
Mauboussin, Michael J., 2012. The Success Equation: mid-cap growth, and large-cap growth.
Untangling Skill and Luck in Business, Sports, and Investing.
Boston, Mass.: Harvard Business Review Press. We prefer active share to other active management
measures such as the Industry Concentration Index or
Rowley, James J., Garrett L. Harbron, and Matthew C. Tufano, the Herfindahl index primarily because active share has a
2017. In Pursuit of Alpha: Evaluating Active and Passive
direct and intuitive economic interpretation. With no
Strategies. Valley Forge, Pa.: The Vanguard Group.
short positions, active share is defined such that it can
Sorensen, Eric H., Keith L. Miller, and Vele Samak, 1998. range from 0% to 100%, where 0% represents perfectly
Allocating Between Active and Passive Management. Financial matching the benchmark and 100% represents a
Analysts Journal 54(5): 18–31. portfolio with no overlap at all. Also, as pointed out by
Cremers and Petajisto (2009), the Industry Concentration
Statman, Meir, 2004. The Diversification Puzzle. Financial
Index is a hybrid measure. sharing features of both active
Analysts Journal 60(4): 44–53.
share and tracking error.
Yeung, Danny, Paolo Pellizzari, Ron Bird, and Sazali Abidin, 2012.
Diversification Versus Concentration … and the Winner Is?
Working Paper Series 18. Sydney, Australia: Paul Woolley
Centre for the Study of Capital Market Dysfunctionality,
University of Technology.

17
 Appendix II. Summary statistics

Figure A-1. Historical analysis summary statistics in non-log form, January 2000–December 2017

a. Fund characteristics

Variable Mean Median 5th percentile 95th percentile

Total number of funds 2,136
Age(years) 15.97 12.84 2.91 42.28
Expense (%) 1.19 1.14 0.62 1.95
Turnover (%) 86.89 61.00 12.00 214.00
TNA (USD millions) 1,079.39 185.84 6.21 4,115.50
NumHold (units) 114.40 75.33 30.00 331.33
SCI (%) 28.56 23.78 8.48 78.52
FCI (%) 42.51 40.45 13.46 73.51
ExcRet 0.38 0.19 (5.89) 7.33

b. Correlation structure

Variable Ln(Age) Expense Turnover Ln(TNA) Ln(NumHold) SCI FCI

Age 1.00
Expense -0.15** 1.00
Turnover -0.10** 0.09** 1.00
TNA 0.29** -0.19** -0.07** 1.00
NumHold 0.02** -0.17** 0.02 0.07** 1.00
SCI -0.04** 0.21** 0.21** -0.06** -0.19** 1.00
FCI -0.18** 0.22** 0.02** -0.10** 0.06** 0.13** 1.00
**Significant at 1%.
Notes: We used quarterly data (January 2000 through December 2017) for the oldest share class of U.S. active equity funds available for sale in the United States with more
than ten holdings and excluding funds-of-funds. (Age) is the number of years since the fund began. Expense ratio (Expense) is the annualized difference between quarterly
gross and net excess returns. Observations for which Expense is negative are omitted from the dataset. Turnover ratio (Turnover) measures the fund’s trading activity and is
reported annually by Morningstar, Inc. Quarterly observations have been computed by linearly interpolating annual observations. (TNA) is the fund’s total asset base, net of
fees and expenses. Observations for which TNA is less than $500,000 are omitted from the dataset. (NumHold) represents the number of stocks held by the fund. Sector
concentration (SCI) is annualized and represents the level of concentration in GICS sectors. Factor concentration (FCI) is annualized and represents the level of concentration
in 9-box Morningstar style factors. Excess return (ExcRet) is the quarterly return in excess of the Russell 3000 Index, before fees. Our data sample presents 19 observations in
which the SCI is higher than 100% because of some funds applying leverage.
Sources: Vanguard calculations, using data from Morningstar, Inc.

18
Appendix III. The impact of number of holdings on net excess returns has varied over time

Figure A-2. Ordinary Least Squares (OLS) coefficient and t-statistic of U.S. funds’ net excess returns against
log-number of holdings, three-year rolling window, January 2000–December 2017

150 6

Ln(NumHold) t-statistic
100
Ln(NumHold)

2
50
0
0
–2

–50 –4

–100 –6
2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018

T-statistic
5% confidence bounds for t-statistic
Coefficient

Notes: We used quarterly data for the oldest share class of U.S. active equity funds available for sale in the United States with more than ten holdings and excluding funds-
of-funds. Observations for which Expense is negative or TNA is less than $500,000 are omitted from the data sample. All control variables are included in the regression. All
independent variables are averaged across the three-year rolling window.
Sources: Vanguard calculations, based on data from Morningstar, Inc.

19
 Vanguard Research
P.O. Box 2600
Valley Forge, PA 19482-2600

Connect with Vanguard® > [Link] © 2018 The Vanguard Group, Inc.
All rights reserved.
Vanguard Marketing Corporation, Distributor.

CFA® is a registered trademark owned by CFA Institute. ISGITO 022019