ArticlePDF Available

Abstract and Figures

Using data from a large German brokerage, we find that individuals investing in passive exchange-traded funds (ETFs) do not improve their portfolio performance, even before transaction costs. Further analysis suggests that this is because of poor ETF timing as well as poor ETF selection (relative to the choice of low-cost, well-diversified ETFs). An exploration of investor heterogeneity shows that though investors who trade more have worse ETF timing, no groups of investors benefit by using ETFs, and no groups will lose by investing in low-cost, well-diversified ETFs.
Content may be subject to copyright.
Abusing ETFs*
Utpal Bhattacharya
1
, Benjamin Loos
2
, Steffen Meyer
3
, and
Andreas Hackethal
4
1
Hong Kong University of Science and Technology,
2
University of Mannheim,
3
Leibniz University
Hannover, and
4
Goethe University Frankfurt
Abstract
Using data from a large German brokerage, we find that individuals investing in pas-
sive exchange-traded funds (ETFs) do not improve their portfolio performance, even
before transaction costs. Further analysis suggests that this is because of poor
ETF timing as well as poor ETF selection (relative to the choice of low-cost, well-
diversified ETFs). An exploration of investor heterogeneity shows that though in-
vestors who trade more have worse ETF timing, no groups of investors benefit by
using ETFs, and no groups will lose by investing in low-cost, well-diversified ETFs.
JEL classification: D14, G11, G28
Keywords: Household finance, ETFs, Security selection, Timing
1. Introduction
One of the most successful financial product innovations of the last 20 years is the
exchange-traded fund (ETF).
1
The first ETF was launched in Canada in 1990. As of
February 2016, 4,479 ETFs had been established, with approximately USD 2.7 trillion in
assets under management (roughly the same size as the global hedge fund industry).
2
* For useful comments and discussions, we thank Tyler Shumway, Dimitris Georgarakos, Craig
Holden, Roman Inderst, Jose Martinez, Andrei Shleifer, Noah Stoffman and Joachim Weber.
Conference and seminar comments at the AEA meetings in Boston, American, Cincinnati, the EFA
meetings in Cambridge, the FIRS conference in Quebec City, the FMA meetings in Nashville, GWU,
Goethe University, Hong Kong University, IIM Ahmedabad, IIM Bangalore, IIM Kolkata, Indiana,
ISB Hyderabad, Lehigh, Minnesota, RSM Rotterdam, University of Melbourne, University of New
South Wales, University of Sydney, University of Technology at Sydney, Vanderbilt, WHU Otto
Beisheim School of Management, Yale and York significantly improved the paper.
1 An ETF is an index-linked security. These instruments aim to replicate the movements of a particu-
lar market and therefore enable the investor to easily buy and sell a broadly diversified portfolio of
securities that mimic that market. Investors can buy and sell ETF shares in public markets any time
during the trading day.
2 ETFGI (Global ETF and ETP Directory, February 2016) and Hedge Fund Research (Global Hedge
Fund Industry Report, Year End 2015).
V
CThe Authors 2016. Published by Oxford University Press on behalf of the European Finance Association.
All rights reserved. For Permissions, please email: journals.permissions@oup.com
Review of Finance, 2016, 1–34
doi: 10.1093/rof/rfw041
Review of Finance Advance Access published August 7, 2016
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
In this paper, we investigate whether or not ETFs provide benefits to a sample of indi-
vidual investors who include them in their portfolios.
3
Given the paucity of studies on the
user effects of financial product innovations like ETFs, this is an important topic to analyze.
Frame and White (2004, p. 116) state: “Everybody talks about financial innovation, but
(almost) nobody empirically tests hypotheses about it.” It is important to test whether ETFs
benefit individual investors because they attract a lot of them.
4
In addition, employers are
actively seeking ways to include ETFs in 401(k) defined-contribution retirement plans
5
and
numerous fin-tech startups promote standardized ETF portfolios to retail investors. Even
some industry regulators are promoting ETFs to individual investors.
6
Our null hypothesis is that individual investors benefit by using passive ETFs. Classical
finance theory prescribes well-diversified and low-cost portfolios for investors.
7
However,
many researchers document substantial portfolio underperformance by individual investors
due to poor diversification and costly over-trading in single stocks.
8
Indeed, ETFs may help
investors attain theoretically sound portfolios.
9
ETFs have other benefits, too. They trade
in real time and they offer tax advantages (Poterba and Shoven, 2002).
However, there is some evidence that investors may not be using index-linked products
wisely. Hortac¸su and Syverson (2004) find large fee dispersions among financially homoge-
neous funds and Elton, Gruber, and Busse (2004) show that investors irrationally prefer more
expensive index funds.
10
Second, it is possible that some ETFs, because they are highly
3 We examine only passive ETFs that aim to mimic an index. Active ETFs, which aim to outperform
an index, are not examined. Among passive ETFs, we do not differentiate whether ETFs are syn-
thetic or fully replicating, despite the fact that synthetic ETFs may entail additional risk
(Ramaswamy, 2011). In unreported analyses, we also look at passive index funds and find results
similar to those for passive ETFs.
4 Charles Schwab, the largest US discount brokerage, offers more than 200 commission-free ETFs
to individual investors (Schwab ETF OneSource, http://www.schwab.com/public/schwab/invest
ing/accounts_products/investment/etfs/schwab_etf_onesource).
5 “Are ETFs and 401(k) Plans a Bad Fit?” The Wall Street Journal, April 5, 2012.
6 The Securities and Markets Stakeholder Group of the European Securities and Markets Authority
(ESMA) states that “ETFs are a low cost and straightforward investment proposition for investors
and, as such, ESMA should investigate how to make indexed ETFs more offered to individual in-
vestors.” ESMA Report and Consultation paper—Guidelines on ETFs and other UCITS issues. July
25, 2012, http://www.esma.europa.eu/system/files/2012-474.pdf, p. 32.
7Markowitz (1952) suggests we diversify by buying optimal portfolios. Tobin (1958) suggests that
we require only one optimal portfolio provided that a risk-free asset exists. Sharpe (1964) con-
cludes that this optimal portfolio was the market portfolio.
8 The portfolios of individual investors who participate in equity markets typically show suboptimal
degrees of diversification (e.g., Blume and Friend, 1975;Kelly, 1995;Goetzmann and Kumar, 2008)
and concentration on the home region (“home bias,” e.g., French and Poterba, 1991;Cooper and
Kaplanis, 1994;Lewis, 1999;Huberman, 2001;Zhu, 2002;Ahearne, Griever, and Warnock, 2004;
Calvet, Campbell, and Sodini, 2007). Individual investors are also shown to trade too much (Odean,
1999;Barber and Odean, 2000).
9Boldin and Cici (2010) review the entire empirical literature on index-linked securities and discuss
their benefits. French (2008) measures the benefits of passive investing and concluded, “the typ-
ical investor would increase his average annual return by 67 basis points over the 1980–2006
period if he switched to a passive market portfolio.”
10 Choi, Laibson, and Madrian (2010) confirmed this behavior in an experiment and found that more
financially sophisticated investors pay lower fees.
2U. Bhattacharya et al.
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
correlated with an index and are easy to trade, may enhance investors’ temptation to time the
underlying index.
11
Third, investors may be overwhelmed by the sheer number of ETF
products and underlying market and sector indices (over 220 such indices in our sample
alone) and end up purchasing costly ETFs linked to rather undiversified single sectors or
industries.
The key contribution of this paper to the literature (to our knowledge, the first of its
kind) is that we use the trading data of a large number of individual investors at a large
German brokerage firm during the 2005–2010 period to test whether ETFs benefit those
who use them.
12
First, we examine who uses ETFs. We find that, compared with non-users,
ETF users are younger, wealthier in terms of both portfolio value and overall wealth, and
have a shorter relationship with the brokerage. Mu¨ ller and Weber (2010), using a survey
methodology, report comparable results.
Second, we compare the portfolio performance of ETF users with all non-users in a
panel setting. We estimate the marginal contribution of ETFs to an individual’s portfolio
performance starting with the first month of ETF use. We examine raw returns as well as
risk-adjusted returns using one, two, four, and five risk factors.
13
We use a panel setting
with user fixed effects to control for any time-invariant differences between users and non-
users of ETFs. We also control for observable demographics, lagged time-varying portfolio
characteristics like prior portfolio performance, and year fixed effects. We find that port-
folio performance, as measured by any of our measures using any benchmark index, does
not increase with ETF use.
Third, and importantly, we examine why there is no performance improvement for ETF
users and what the performance improvement would have been had investors used ETFs
wisely. The basic idea is to compare actual portfolios with counterfactual portfolios. This
approach allows for inferences at the individual investor level, mitigating issues of self-
selection and endogeneity.
We start with our first benchmark portfolio that is the non-ETF part of the portfolio. If
we add all the actual ETF trades of an investor, we are back at the full portfolio. The return
differential between the benchmark portfolio and the full portfolio is a statistically signifi-
cant 1.16% per year. We then examine what would happen if the actual ETFs were only
bought, but not sold, essentially emulating an ETF buy-and-hold strategy. This counterfac-
tual portfolio allows one to extract the contribution coming from ETF timing ability. We
find that poor ETF timing ability is responsible for 0.77 percentage points (statistically
significant) of the total return differential (1.16%). ETF selection ability is not statistically
11 In Germany, by 2009, the turnover in ETFs (data obtained from Deutsche Bo¨ rse 2010) had become
about the same as the turnover in stocks (data obtained from the World Federation of Exchanges
2013).
12 We test whether the portfolio performance of individual investors improves after they purchase
ETFs. An ex ante test like the one proposed by Calvet, Campbell, and Sodini (2007) will fail to in-
corporate the dynamic effects of actual trading.
13 For the market factor, we use a global index (MSCI All Country World Index “MSCI ACWI”), as
well as the broadest local index (CDAX). The former benchmark is for global investors and the lat-
ter benchmark is for local investors. We use both indices for robustness. In our factor models that
include a bond factor, we add the JP Morgan Global Bond to the MSCI ACWI as a fixed income
benchmark for global investors and the RDAX Return Index to the CDAX Index as a fixed income
benchmark for local investors.
Abusing ETF 3
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
significant. Examining gross returns and risk-adjusted gross returns confirms that the actual
portfolio returns of ETF users are mainly adversely affected by poor ETF timing, though
trading costs matter as well. Focusing on portfolio efficiency alone, we find that the relative
Sharpe ratio loss (RSRL)
14
increases significantly with ETF use. This rules out that investors
use ETFs mainly for hedging or better diversification.
Our second benchmark portfolio is a prescription: we prescribe the investor a buy-
and-hold strategy in a low-cost ETF on the MSCI World Index. We find that investors are
losing a statistically significant 1.69% p.a. in net portfolio returns by not using this pre-
scribed portfolio. To decompose the above loss, we start with the actual portfolio of the
investor. We then examine what happens if we replace all ETF trades with trades in a
low-cost ETF on the MSCI World Index. This particular counterfactual portfolio isolates
ETF selection (relative to choosing the MSCI). We find that most of that 1.69% loss
(1.28%, statistically significant) would have come from ETF selection (not choosing the
low-costETFontheMSCIWorldIndex)andlittle(0.41%, not statistically significant)
from not employing a buy-and-hold strategy. This result also holds for gross portfolio re-
turns, gross risk-adjusted returns, and diversification. We conclude that the average in-
vestor could have benefited from using ETFs by following the guidelines of classical
finance theory.
Finally, we explore investor heterogeneity in terms of overconfidence (proxied by port-
folio turnover) and financial sophistication (proxied by portfolio value and portfolio diver-
sification) to see if there are specific types of investors where our results are most relevant.
Our conclusion from sorting investors by overconfidence and sophistication: though in-
vestors who trade more have worse ETF timing, no groups of investors benefit by using
ETFs, no matter which measure (performance, timing, selection, or diversification) or sort
(turnover, portfolio value, or diversification) we examine. We also find that no groups will
lose by investing in the right MSCI ETF.
Our sorting exercise also yields one potential explanation. Investors from virtually all
groups do not substantially adapt their trading behavior after ETF use. Those who traded
more before ETF use continue to trade more after ETF use, both in the ETF part of the port-
folio, as well as in the non-ETF part. Investors therefore appear to make the same mistakes
when they trade ETFs that they have made in trading non-ETFs.
Our overall conclusion is that our sample of ETF users does not improve their actual
portfolio performance after ETF use because they have both poor ETF timing as well as
ETF selection (relative to choosing a low-cost well-diversified ETF like the MSCI). Thus, al-
though passive ETFs are an important investment innovation, with an enormous potential
to act as a low-cost vehicle for diversification, in practice they may not help individual in-
vestors enhance the efficiency of their portfolio, even before transaction costs. This would
happen if individual investors get tempted to trade too much in the ever-expanding choices
of high-liquidity ETFs based on narrow market indices. To conclude, more ETF choice may
lead to abuse of ETFs.
We describe the data in Section 1 and examine which investors are most likely to pur-
chase ETFs in Section 2. In Section 3, we investigate whether ETF users improve their port-
folio performance compared with non-users. In Section 4, we examine why ETF users do
not improve their relative portfolio performance. We conclude in Section 5.
14 We measure the relative Sharpe ratio loss as defined in Calvet, Campbell, and Sodini (2007).
4U. Bhattacharya et al.
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
2. Data
2.1 ETFs and Index-Linked Securities in Germany
Individuals in Germany, as in the USA, who want to invest in index-linked securities can
choose ETFs and/or index mutual funds. Table I gives us a snapshot of both markets at the
end of a year. Panel A of Table I provides the data for index-linked securities in Germany.
Panel B provides this information for the USA. Panel C provides the data for our German
sample. As a result of data availability, the three panels represent a snapshot of the market
at different times. For Germany and the USA, the data for the end of 2011 are available,
whereas these data for our sample are available only for the end of 2009.
The leftmost column in Panels A and B of Table I shows that the total assets under
management invested in index-linked securities relative to total active mutual fund invest-
ments, a ratio of about 20%, is comparable between Germany and the USA. Panels A and
B also show that the market in the USA, as expected, is much larger as measured by both
assets under management and the number of products. Interestingly, in terms of assets
under management, the market is split almost evenly between passive ETFs and index
mutual funds in the USA, whereas in Germany, passive ETFs comprise 84% of the
market.
When Panel A of Table I (Germany) is compared with Panel C (our sample), in terms of
the proportion of assets under management in each security class, our sample seems to be
representative of the entire German market.
Table I. Usage of index-linked securities: an overview
This table provides an overview of the markets for ETFs and index funds in Germany (Panel A),
the USA (Panel B), and within our sample (Panel C). For all panels, the latest available year-end
data are used. We report the number of products, as well as assets under management (AUM),
in absolute numbers and in percentages. The last two columns show the ETFs and index funds
with active mutual funds in terms of the number of available products and AUM.
Index-linked securities As % of active mutual funds
Number of products % AUM in em % Number of products AUM
Panel A: Index-linked securities in Germany
a
Passive ETFs 826 86% 99,311 84%
Index mutual funds 135 14% 18,353 16%
Total 961 100% 117,664 100% 17% 20%
Panel B: Index-linked securities in the USA
b
Passive ETFs 1,028 73% 934,216 46%
Index mutual funds 383 27% 1,094,296 54%
Total 1,411 100% 2,028,512 100% 23% 21%
Panel C: Index-linked securities held by our investors
c
Passive ETFs 279 90% 17 95%
Index mutual funds 30 10% 1 5%
Total 309 100% 18 100% 17% 16%
a
As of December 31, 2011. Source: BVI, Deutsche Bo¨ rse.
b
As of December 31, 2011. Source: Investment Company Institute Factbook 2012.
c
As of December 31, 2009.
Abusing ETF 5
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
2.2 ETFs in Our Sample
In this paper, we focus only on ETFs rather than index funds for two reasons. First, as can
be seen in Table I, ETFs are the predominant index-linked security in Germany, as well as
in our sample. Second, as the construction and trading of index funds are different from
ETFs, we do not bundle the two.
15
Table II shows the rich diversity of ETFs in the portfolios in our sample. Panel A shows
that our investors have exposure to many different indices. Although the top 10 benchmark
indices constitute over 65% of the assets under management in ETFs, 224 other benchmark
indices make up the remainder. Note that the popular indices are connected to Germany,
Europe, and the world, which motivates us to select the local German index, CDAX, and a
global index, MSCI ACWI, as our two benchmark indices.
In Panel B of Table II, we examine the regional allocations of these ETFs. Europe is the
most popular, followed by Germany. Individual German investors, like individual investors
all over the world, exhibit home bias.
In Panel C of Table II, we examine the asset class of ETFs. We find that ETFs that are
based on equity indices dominate (90.5% of assets under management), which further justi-
fies our use of equity indices like CDAX and MSCI ACWI as benchmarks. However, as
there are a few bond- and commodity-based ETFs as well, we will sometimes use a bond
benchmark.
Panel A of Table II shows that many ETFs in our sample are linked to narrow indices,
so it is likely that they offer more choices for timing certain asset classes, sectors, or coun-
tries, rather than opportunities for broad diversification. If so, their beta loadings with re-
spect to our benchmarks, CDAX and MSCI ACWI, could be very different from 1. In Panel
DofTable II, we show the beta loadings of all ETFs in our sample with respect to the
CDAX and the MSCI ACWI. The mean beta loadings with respect to the CDAX and the
MSCI ACWI are 0.72 and 0.88, respectively. Although these betas are statistically signifi-
cantly different from 1, if we narrow our sample to equity ETFs, the mean beta loading
with respect to the MSCI ACWI cannot be distinguished from 1, but the mean beta loading
with respect to the CDAX is still below 1. Further, although many of these ETFs may not
be tracking the CDAX or the MSCI ACWI perfectly, Panel D of Table II shows that their
alphas with respect to these indices are indistinguishable from zero.
2.3 Individual Investors in Our Sample
The brokerage that we work with was founded as a direct bank with a focus on offering
brokerage services via telephone and the Internet. In 2009, to retain existing customers and
attract new ones, the brokerage introduced a financial advisory service, which offered free
financial advice to a random sample of about 8,000 investors. Approximately 96% of these
individuals refused the financial advice and continued trading as before.
16
Our starting
sample is these 7,761 investors. The knowledge that these investors refused financial advice
assures us that our sample is composed of self-directed investors whose decisions are not
15 The economic intuition of our paper, however, applies to both index funds and ETFs. Therefore, as
mentioned in footnote 3, we replicate all our tests for passive index funds. We find results similar
to those for passive ETFs.
16 Bhattacharya et al. (2012) analyze the same sample with a focus on the 4% of individual investors
who accepted the offer.
6U. Bhattacharya et al.
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
Table II. The kind of ETFs investors in the sample buy
Panel A: This panel shows the average amount of euros invested per month in a passive ETF on
a benchmark index as a percentage of the total average amount of euros invested per month in
all passive ETFs. Panel B: This panel shows the average amount of euros invested per month in
a region using passive ETFs as a percentage of the total average amount of euros invested per
month in all passive ETFs. Panel C: This panel shows the average amount of euros invested per
month in an asset class using passive ETFs as a percentage of the total average amount of
euros invested per month in all passive ETFs. Panel D: This panel shows the distribution of
beta, alpha, and tracking error of all ETFs (top panel) and ETFs based on equity indices (bottom
panel) that investors in our sample use. Beta, alpha, and tracking error (RMSE) result from a re-
gression of ETF returns on the MSCI ACWI or the German benchmark index CDAX and are esti-
mated separately for each ETF. p-Values result from a t-test of betas and alphas against 1 and 0,
respectively. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels,
respectively.
Panel A
Benchmark index Share in %
DAX 25.0%
STOXX Europe 50 11.2%
STOXX Europe Select Dividend 5.7%
LevDAX 4.2%
MDAX 3.7%
TecDAX TRI 3.7%
MSCI World 3.3%
EONIA 3.2%
MSCI Emerging Markets 3.0%
STOXX Europe 600 2.5%
Other (224 indices) 34.4%
Total 100.0%
Panel B
Country/region Share in %
Europe 42.1%
Germany 35.6%
Emerging markets 6.2%
World 3.5%
Japan 3.2%
USA 2.6%
China 1.8%
Brazil 1.2%
India 0.9%
Asia 0.8%
Other 2.2%
Total 100.0%
(continued)
Abusing ETF 7
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
distorted by a third party. As our focus is on ETFs, we keep investors who invest in all
securities except index mutual funds. We additionally restrict our sample to investors who
on average have at least e5,000 in their portfolios. We do so to avoid a bias introduced by
small play money accounts. Our final sample has 6,949 investors in an unbalanced panel
that begins in August 2005 and ends in March 2010. Of these 6,949 investors, 1,080 in-
vestors traded at least one ETF during this period—the “users”—and 5,869 investors who
traded no ETFs during this period—the “non-users.”
Figure 1 shows the share of ETFs in the portfolio of an average individual investor in
our sample. It shows that after investors have switched to ETFs, their weight in the port-
folio hardly exceeds 20%. Figure 1 also shows the growing popularity of ETFs in our sam-
ple. The sharp increase in ETF share in December 2008 is likely related to a tax change in
Germany. From 2009 onwards, all capital gains and losses, irrespective of the holding
Table II. Continued
Panel C
Asset class Share in %
Equity 90.5%
Bonds 6.5%
Commodities 3.0%
Other 0.1%
Total 100.0%
Panel D
All ETFs
Metric NMean p-value Median 10% 25% 75% 90%
Benchmark: MSCI World All Country
Beta 353 0.88 0.002*** 1.03 0.04 0.61 1.29 1.59
Alpha in % p.a. 353 0.84 0.237 0.25 12.27 4.36 5.17 12.99
Tracking Error in % p.a. 353 3.63 3.16 1.27 2.16 4.70 6.21
Benchmark: CDAX
Beta 353 0.72 0.000*** 0.83 0.07 0.44 1.09 1.37
Alpha in % p.a. 353 0.11 0.237 0.36 13.11 6.30 5.66 13.41
Tracking Error in % p.a. 353 3.55 3.09 1.13 2.11 4.59 6.34
ETFs on equity indices
Metric NMean p-value Median 10% 25% 75% 90%
Benchmark: MSCI World All Country
Beta 284 1.05 0.238 1.11 0.60 0.91 1.35 1.66
Alpha in % p.a. 284 0.67 0.439 0.07 13.06 4.84 5.45 14.21
Tracking Error in % p.a. 284 3.90 3.31 1.81 2.41 4.81 6.34
Benchmark: CDAX
Beta 284 0.87 0.000*** 0.93 0.44 0.69 1.15 1.40
Alpha in % p.a. 284 0.17 0.837 2.03 13.86 6.67 5.95 13.91
Tracking Error in % p.a. 284 3.79 3.26 1.70 2.40 4.72 6.38
8U. Bhattacharya et al.
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
period, are subject to taxation. Gains and losses from securities purchased before the end of
2008, if held for longer than 1 year, are tax exempt. Thus, it is possible that some investors
switched to ETFs in December 2008 to ensure a tax advantage.
The German brokerage provided us with investor demographics and account characteris-
tics for both ETF users and non-users for the sample period. Investor demographics include
gender, age, and micro-geographic status. The micro-geographic status variable measures the
average wealth level of individuals who inhabit a given micro area (street-level address). The
variable has nine categories, with category one comprising the poorest individuals and cat-
egory nine the wealthiest individuals. This information is provided to the German brokerage
by a specialized data service that uses several factors (such as house type and size, dominant
car brands, rent per square meter, and the unemployment rate) to construct it.
The account characteristics are primarily comprised of monthly position statements,
daily transaction data, and account transfers for the August 2005 to March 2010 period.
We use the transaction records to calculate portfolio turnover and number of trades per
month, as in Barber and Odean (2002). To compute daily position statements and portfolio
values, we proceed as follows. We multiply the beginning-of-month value of each security
holding by the corresponding daily price return (excluding dividends but considering any
capital actions) for that security to obtain its end-of-day holding value. These values are
then adjusted for any sales, purchases, and/or account transfers that occurred on that day
to yield the position statements for the beginning of the second day in the month. We repeat
this procedure for each trading day in a given month. The computed holdings on the last
day of each month are then reconciled with the true holdings in our dataset.
Daily portfolio returns are calculated as the weighted average return of all securities
held, purchased, and sold by the investor on that day. For securities held, we use total daily
Figure 1. ETF use in our sample. The figure presents the usage of ETFs over time. The solid line shows
the average percentage share of ETFs in terms of euros in the portfolios of users (ETF share in %) and
the dashed line shows the cumulative number of users (Number of users of ETFs).
Abusing ETF 9
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
return data from Datastream (they take into account dividend payments). For securities
that are either purchased or sold on that day, we compute daily returns based on exact
transaction prices. Our weighting factors for securities held or sold are closing prices of the
previous day times the number of securities held or sold. The weighting factors for secur-
ities purchased are the corresponding transaction prices multiplied by the number of secur-
ities purchased. Since we also obtained transaction costs, commissions, and fees from the
bank, we are able to calculate daily security and portfolio returns both on a gross (before
transaction cost) and on a net (after transaction cost) basis.
The account characteristics provided by the brokerage also include account opening date
and cash account balances at the beginning of the sample period and at the end of the sample
period. The account opening date gives us the length of the client relationship with the
brokerage, and the cash account balances enable us to calculate the share of risky securities in
the account with the brokerage (portfolio value plus cash value) for at least two dates.
Table III gives the sources of all the data described above, as well as of data obtained
from other sources. Finally, as we find that the typical investor in our sample only trades
about twice a month, we aggregate all daily returns and other statistics to the monthly
level.
3. Who Uses ETFs?
Table IV provides summary statistics about the users and non-users of ETFs in the sample.
In this univariate setting, ETF users seem to be slightly younger and wealthier than non-
users. Moreover, they also have a shorter relationship with this brokerage, a higher share of
Table III. Data collected
The data are summarized in this table.
Type of data Data Frequency Source of data
Client
demographics
Gender Time-invariant Bank
Date of birth (measure of age) Time-invariant Bank
Microgeographic status (measure of
wealth)
Time-invariant Bank
Portfolio
characteristics
Actual position statements Monthly Bank
Actual transactions and transfers Daily Bank
Cash On start and
end of dataset
Bank
Account opening date (measure of
length of relationship)
Time-invariant Bank
Market data German Fama and French (1993)
and Carhart (1997) factors
Daily Datastream/
own calculation
MSCI World All Country Daily Datastream
CDAX Daily Datastream
RDAX Daily Datastream
JP Morgan Global Bond Daily Datastream
Individual security prices Daily Datastream
Individual security properties Time-invariant Bank/Deutsche
Bo¨ rse
10 U. Bhattacharya et al.
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
Table IV. Summary statistics
This table reports summary statistics on investor demographics, investor characteristics, and portfolio characteristics. Investor demographics are comprised of
statistics on the share of male investors (Gender), the age of investors (Age), and the wealth of an investor measured by the micro-geographic status rating, one
through nine, assessed by an external agency (Wealth). Investor characteristics are comprised of statistics on the number of years the investor has been with
the bank (Length of relation) and the proportion of risky assets (Risky share) held with this brokerage at the beginning (08/2005) and at the end (03/2010) of our
sample period. Portfolio characteristics are comprised of the following statistics: the average risky portfolio value (Average risky portfolio value) of the customer,
the average number of securities in the portfolio at the end of each month (Average number of securities), the average number of trades per month (Average
number of trades), the average portfolio turnover per month (Average portfolio turnover), and alphas net of transaction costs for the MSCI ACWI and the CDAX
(alpha). ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.
ETF users ETF non-users t-Test (users vs.
non-users)
Metric Measurement units Mean Median NMean Median Np-value
Client demographics
Gender Dummy ¼1 if male 80.9 100.0 1,080 82.0 100.0 5,869 0.405
Age (08/2005) Years 47.8 45.0 1,080 49.8 48.0 5,869 0.000***
Wealth (08/2005) Microgeographic status 6.5 7.0 952 6.3 6.0 5,164 0.015**
Investor characteristics
Length of relationship with the bank (03/2010) Years since account opening 7.2 8.1 1,080 7.6 8.9 5,869 0.000***
Risky share (08/2005) % 81.6 85.5 754 95.5 86.1 4,418 0.653
Risky share (03/2010) % 78.0 86.7 1,043 73.4 82.2 5,381 0.000***
Portfolio characteristics
Average risky portfolio value (08/2005–03/2010) ethousands 60.3 42.8 1,080 51.0 34.6 5,869 0.000***
Average number of securities (08/2005–03/2010) Count 12.0 9.7 1,080 10.9 8.5 5,869 0.001***
Average number of trades (08/2005–03/2010) Trades per month 2.1 1.4 1,080 1.6 0.9 5,869 0.000***
Average portfolio turnover (08/2005–03/2010) %, monthly 7.4 4.4 1,080 6.5 3.5 5,869 0.001***
Alpha (net) MSCI World All Country (08/2005–03/2010) %, yearly 0.9 0.0 1,080 2.1 0.4 5,869 0.091*
Alpha (net) CDAX (08/2005–03/2010) %, yearly 3.2 2.3 1,080 3.9 2.8 5,869 0.300
Abusing ETF 11
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
their portfolio in risky securities at the end of the sample period, a higher average portfolio
value during the sample period, more securities in their portfolio, and they trade more often
during the sample period. We also find a small difference in alpha over the entire sample
period, suggesting that ETF users appear to be more skilled investors than the non-users.
Table V provides the results of a multivariate probit model to confirm the above univari-
ate results. The dependent variable is set to one if an investor opted to use ETFs at least
once in our sample period and is set to zero otherwise. The independent variables are the
time-invariant variables that we know at the start of our sample or on the first day an in-
vestor enters the sample. The results in Table V confirm that younger and wealthier (in
terms of portfolio value) investors are more likely to use ETFs. This echoes the survey re-
sults in Mu¨ ller and Weber (2010) and is consistent with findings in the marketing literature
(e.g., Dickerson and Gentry 1983) that document early adopters to be younger and
wealthier.
Table V. ETF users: a probit test
This table reports the marginal effects of a probit regression. The dependent variable is a
dummy (Dummy user) set to one for individual investors who hold at least one ETF during the
sample period. For the estimation of the probit model, our independent variables are time-in-
variant or measured at the beginning (08/2005) of our sample period or at the first day an in-
vestor enters our sample. The independent variables are: a dummy that is equal to 1 if an
investor is male (Dummy male), the age of an investor (Age), a dummy that is equal to 1 if an in-
vestor falls into categories 1–3 of a micro-geographic status rating (Dummy low wealth), a
dummy that is equal to 1 if an investor falls into categories 7–9 of the micro-geographic status
(Dummy high wealth), the risky portfolio value in euros of the investor (Log portfolio value) on
the day he enters the sample, years the investor has been with the bank (Length of relation),
and the proportion of risky assets in the account (Risky share) on the day the investor enters
the sample. Heteroscedasticity robust p-values are in parentheses. ***, **, and * denote statis-
tical significance at the 1%, 5%, and 10% levels, respectively.
Dummy user
(1) (2) (3) (4)
Dummy male 0.011 0.012 0.006 0.013
(0.328) (0.313) (0.607) (0.351)
Age (08/2005) 0.002*** 0.002*** 0.002*** 0.002***
(0.000) (0.000) (0.000) (0.000)
Dummy low wealth (08/2005) 0.008 0.008 0.009 0.011
(0.670) (0.678) (0.642) (0.602)
Dummy high wealth (08/2005) 0.015 0.014 0.014 0.005
(0.107) (0.112) (0.129) (0.626)
Log portfolio value (first day) 0.004 0.007** 0.015***
(0.114) (0.013) (0.000)
Length of relationship (08/2005) 0.005*** 0.001
(0.001) (0.575)
Risky share (08/2005) 0.000
(0.593)
Observations 6,949 6,949 6,949 5,172
Pseudo-R
2
0.00470 0.00516 0.00715 0.00657
12 U. Bhattacharya et al.
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
4. Do Individual Investors Benefit by Using ETFs?
We now examine whether individual investors benefit by using ETFs. We use data from all
ETF users and non-users in our sample. This allows us to exploit all of the information in
our panel dataset. We thus estimate the following model:
ri;t¼/þb1First Use of ETFsi;tþb2User fixed effect iþb3MFtþb4TFt
þb5DCiþb6ICi;t7to t1ðÞ
þj;(1)
where ri;tis the excess net return (excess over the 3-month Euribor rate and net of all trans-
action cost) on investor i’s portfolio in month t,adenotes the constant, First Use of
ETFsi;tis a dummy variable set to 1 in every month tafter investor ihas invested in ETFs
for the first time, User fixed effect is a dummy variable set to 1 if an investor holds an ETF
at any point in time during our observation period, and MFtis a vector representing the re-
turn of factors like the market factor in month t. Depending on the specification, this vector
may contain no factors, a market factor (CDAX or MSCI ACWI) or additional factors like
SMB (small-minus-big), HML (high-minus-low), MOM (Momentum), or a bond factor.
TFtrepresents year fixed effects, which means that there is one year dummy for each year.
DCiis a demographic control vector for investor i. This vector contains gender, age, dum-
mies for low and high wealth, and length of relationship. ICi;t7to t1ðÞ
is a vector of time-
varying characteristics (log of the portfolio value, alpha, turnover, and number of trades) of
the portfolio of investor iover the rolling window t7 (months) to t1. All these time-
varying portfolio characteristics of the investor are rolling moving averages calculated on a
monthly basis at tover the prior 6 months from t7tot1 (6 months MA).
The use of year fixed effects is important in our context. These control for any events in
a given calendar year that change the propensity to purchase ETFs, such as the financial cri-
sis years of 2007–2009 or years in which the tax policy on investment profits changed. In
our sample, this is particularly important since a tax law change took place in Germany at
the end of 2008. From 2009 onwards, all capital gains and losses, irrespective of the hold-
ing period, are subject to taxation. Gains and losses from securities purchased before the
end of 2008, if held for longer than one year, are tax exempt. Because some investors may
have purchased ETFs to ensure a tax advantage in 2009 (Figure 1), a year with above aver-
age stock returns, the effect of buying ETFs for tax reasons in this year would indicate a
spurious benefit of ETF use.
Although the user fixed effects control for all time-invariant differences in characteris-
tics of users and non-users of ETFs, the criticism remains that the choice of using an ETF
may still be endogenous because we have not controlled for time-varying variables. To miti-
gate this concern, we control for the following time-varying portfolio characteristics of the
investor that we can observe: log portfolio value, past performance as measured by a one-
factor Jensen (1968) alpha with the CDAX as the benchmark, and trading behavior meas-
ured by number of trades and portfolio turnover. We use the rolling moving average of the
previous six trading months to calculate these four variables.
Finally, when running these panel regressions, we cluster standard errors by month in all
the regressions to address potential issues with cross-sectional correlation (Seasholes and
Zhu, 2010) and to be as conservative as possible. This level of clustering leads to lower t-stats
than a two-way cluster on investor and month, as suggested by Petersen (2009).Ifwehad
not clustered standard errors by month, and would therefore have assumed independence of
returns between investors, the t-statistics would have been on average five times higher.
Abusing ETF 13
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
The independent variable of interest is First Use of ETFs. It is set to 0 on months before
ETFs were used, and switched to 1 after the first use of ETFs no matter whether investors
held any ETFs in subsequent periods. This allows us to compare portfolios of users before
and after the use of ETFs. So b1measures the change in portfolio performance after an in-
vestor uses ETFs for the first time. If we run Equation (1) without MFt, the coefficient b1
measures the change in portfolio performance without risk adjusting, whereas if we run the
equation with MFt, the coefficient measures the change in portfolio performance after risk
adjustment. The b3coefficients are the corresponding betas or factor sensitivities. The vari-
able User fixed effect, which is set to 1 if an investor holds an ETF at any point in time dur-
ing our observation period, allows us to compare ETF users with non-users. Their
differential portfolio performance is measured by b2. The b4coefficients are the fixed ef-
fects of a given year, which we do not show in Table VI for the sake of brevity. The b5coef-
ficients are the effects of the investor’s time-constant demographic characteristics. The b6
coefficients are the effects of the investor’s time-varying portfolio characteristics.
Although the above specification seems like a classic difference-in-difference research
design, it is not in our context. The reason is that it is not clear what the exact treatment
and control groups are. Certainly, investors who have never held an ETF in our sample are
unequivocally non-users and belong to the control group and investors who purchase ETFs
for the first time and then keep ETFs in their portfolio over the remaining sample period
clearly belong to the treatment group. However, if an investor held an ETF in the past but
does not hold an ETF in month t, should she be assigned to the control group of non-users
or the treatment group of users for month t? Questions like these are important, and it is
for this reason that we run Equation (1) in many ways.
17
However, given the lack of an ex-
ogenous shock, the myriad possible ways of running our panel regressions or doing a pro-
pensity score matching, will still not give us a clean identification. Recognizing this
limitation, and noting that the results we obtain from our various tests are qualitatively
similar, we show the results of just one of our tests in Table VI.
Column (1) in Table VI gives the results for raw net returns, whereas the other columns
give the results for risk-adjusted net returns. We risk adjust using the one-factor MSCI
ACWI, the MSCI ACWI factor and a world bond factor, the one-factor CDAX, the CDAX
factor, and a German bond factor, the four-factor model that uses the CDAX factors, and
the five-factor model that uses the CDAX factors and a German bond factor, and present
the results in Columns (2)–(7), respectively.
The most important result in Table VI is the observation that the portfolio performance of
ETF users does not improve after they start using ETFs; the coefficient on the First Use of
17 We re-run regression (1) by restricting our sample only to users, and we define First Use of ETFs
only for investors who hold ETFs every month after first use or define First Use of ETFs only for in-
vestors who hold ETFs sometimes after first use. Results are similar. Results are qualitatively un-
altered if we add non-users to the sample and re-run the above two tests. Results are
qualitatively unaltered if we allow the factor exposures to be different for ETF users and non-
users. The logic is that ETFs, being a basket of well-diversified securities, may tend to move the
beta of the portfolios toward 1. Our results are qualitatively similar if we use the continuous frac-
tion of an investor’s portfolio value that is invested in ETFs instead of the dummy specification.
Results are similar if, instead of using the user fixed effect and demographic control variables as
in Table VI, we use investor fixed effects. We get the same qualitative result using a propensity
scoring methodology. All results are available from the authors upon request.
14 U. Bhattacharya et al.
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
Table VI. Do ETFs improve portfolio performance?
This table reports the results of a panel regression where the dependent variable is the net return of an investor (model 1) or excess net return of an investor
(models 2–7). Here, excess net return is the excess return over the 3-month Euribor rate. The sample includes all non-users and all users of ETFs. The independ-
ent variable of interest is First Use of ETFs, which is set to 1 from the first month in which an investor holds an ETF. We also include a user fixed effect usinga
dummy variable (User fixed effect), which is set to 1 if an investor holds an ETF during our sample period. The regressions include the following independent
variables: time-varying risk factors (MSCI ACWI, World Bond, CDAX, German Bond, CDAX (SMB), CDAX (HML), and/or CDAX (MOM)), and time-varying port-
folio characteristics of the investor (the log of the risky portfolio value in euros (Log portfolio value), the systematic risk-adjusted return (Alpha), portfolio turn-
over, and average number of trades). All these time-varying portfolio characteristics of the investor are rolling moving averages calculated on a monthly basis
at tover the prior 6 months from t7tot1 (6 months MA). “Alpha (net) 6 months MA” comes from a regression of excess net portfolio returns on the German
benchmark index CDAX in the t7tot1 window, and is estimated separately for each investor. The regression is estimated with demographic controls as well
as year fixed effects. Demographic controls, defined in Table IV, are: gender, age, dummy low and high wealth, and length of relation. Standard errors are clus-
tered by month. P-values are in parentheses. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.
Performance
(1) (2) (3) (4) (5) (6) (7)
Net return Excess net
return
Excess net
return
Excess net
return
Excess net
return
Excess net
return
Excess net
return
First use of ETFs 0.687 0.628 1.358 3.777 4.144 3.380 3.774
(0.887) (0.895) (0.734) (0.222) (0.174) (0.276) (0.217)
User fixed effect 1.439 0.716 0.367 3.033 3.227 2.830 3.032
(0.594) (0.820) (0.890) (0.173) (0.130) (0.197) (0.155)
MSCI World All Country excess return 1.072*** 0.989***
(0.000) (0.000)
World bond index excess return 0.822***
(0.000)
CDAX excess return 0.837*** 0.807*** 0.918*** 0.911***
(0.000) (0.000) (0.000) (0.000)
(continued)
Abusing ETF 15
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
Table VI. Continued
Performance
(1) (2) (3) (4) (5) (6) (7)
Net return Excess net
return
Excess net
return
Excess net
return
Excess net
return
Excess net
return
Excess net
return
German bond index excess return 0.670** 0.656***
(0.011) (0.008)
CDAX (SMB) 0.282*** 0.263***
(0.002) (0.003)
CDAX (HML) 0.035 0.024
(0.756) (0.818)
CDAX (MOM) 0.012 0.062
(0.820) (0.327)
Dummy male 0.014 0.014 0.014 0.014 0.014 0.015 0.014
(0.521) (0.594) (0.588) (0.601) (0.608) (0.585) (0.596)
Age (08/2005) 0.965 1.009 0.944 1.136 1.147 1.137 1.152
(0.345) (0.422) (0.462) (0.375) (0.370) (0.376) (0.369)
Dummy low wealth (08/2005) 0.743 0.789 0.765 0.803 0.808 0.784 0.784
(0.350) (0.402) (0.427) (0.412) (0.413) (0.424) (0.427)
Dummy high wealth (08/2005) 0.485 0.510 0.489 0.509 0.503 0.504 0.501
(0.337) (0.405) (0.438) (0.424) (0.433) (0.430) (0.436)
Length of relationship (08/2005) 0.086 0.126 0.048 0.206 0.211 0.185 0.200
(0.720) (0.491) (0.790) (0.173) (0.152) (0.227) (0.171)
Log portfolio value 6 months MA 4.395*** 3.356*** 3.047*** 2.451*** 2.227*** 2.297*** 2.241**
(0.004) (0.000) (0.001) (0.003) (0.009) (0.006) (0.010)
Alpha (net) 6 months MA 1.308*** 0.929*** 0.898*** 0.896*** 0.896*** 0.867*** 0.873***
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
(continued)
16 U. Bhattacharya et al.
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
Table VI. Continued
Performance
(1) (2) (3) (4) (5) (6) (7)
Net return Excess net
return
Excess net
return
Excess net
return
Excess net
return
Excess net
return
Excess net
return
Portfolio turnover 6 months MA 28.931 12.062 21.323 12.400 13.884 12.282 12.304
(0.228) (0.448) (0.128) (0.288) (0.248) (0.254) (0.264)
Average number of trades 6 months MA 0.450 0.568 0.717* 0.476 0.508 0.519 0.533
(0.402) (0.185) (0.071) (0.245) (0.209) (0.185) (0.173)
Constant 5.825 1.035 7.104 9.711* 8.119 5.101 5.080
(0.645) (0.848) (0.214) (0.065) (0.114) (0.303) (0.213)
Observations 284,866 284,866 284,866 284,866 284,866 284,866 284,866
R-squared 0.194 0.435 0.461 0.476 0.480 0.482 0.485
Number of investors 6,893 6,893 6,893 6,893 6,893 6,893 6,893
Year fixed effect Yes Yes Yes Yes Yes Yes Yes
Abusing ETF 17
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
ETFs is negative in five models and positive in two models, but statistically insignificant in all
seven models. Table VI also shows that ETF users do no worse than non-users over the whole
sample; the coefficient on the User fixed effect is positive but insignificant in each of the seven
models. We interpret both these results to mean that although ETF users are insignificantly
better investors than non-users, ETFs do not improve their portfolio performance after use.
We cannot completely control for unobserved heterogeneity. This is because we do not
have an exogenous shock (even the “exogenous” tax law change in Germany toward the
latter part of our sample period may affect ETF users differently in unpredictable ways) or
good instrumental variables to further refine our testing. We can rule out, however, one
usual suspect—investors using ETFs use all products sub-optimally, not just ETFs—from
the results in both Table IV (alphas are higher for ETF users) and Table VI (ETF users do
no worse than non-users in their portfolio performance as seen in the non-negative coeffi-
cient of the user fixed effect).
5. Why Individual Investors Do Not Benefit by Using ETFs?
5.1 Counterfactual Portfolios
We have shown above that individual investors do not benefit when they hold ETFs in their
portfolios. In this section, we use counterfactual portfolio analysis to determine why.
The basic idea is to compare the performance of actual portfolios with hypothetical port-
folios where we let the investor use a buy-and-hold ETF strategy (in this counterfactual port-
folio, we switch off ETF timing) or we let the investor replace all his ETF buys and sells at time
twith buys and sells in a MSCI World Index ETF at that same time t(in this counterfactual
portfolio, we switch off security selection), or we let the investor use a buy-and-hold MSCI
World Index ETF strategy (in this counterfactual portfolio, we switch off market timing and
security selection). These counterfactual portfolios can be constructed because we know for
each trade of each investor, the ISIN, the date, the value, and the associated fees of that trade.
The counterfactual approach allows for inferences at the individual investor level, miti-
gating issues of self-selection and endogeneity. This is because we look at the portfolio per-
formance changes for the same investor at the same point in time. Thereby, we can directly
draw conclusions on how a change in trading strategy or a different ETF selection changes
individual performance. In contrast to other approaches, we do not have to rely on a single
return series at the portfolio level to decompose portfolio returns into the components of se-
curity selection and market timing.
18
In comparison to Odean (1999), who uses individual
investors’ trading records to decompose the holding period returns of security purchases
and sales into market timing and security selection abilities, our counterfactual portfolio
approach does not require any assumptions on the weighting of trades or the lengths of
holding periods. Risk adjustment is also possible.
19
18 See Treynor and Mazuy (1966),Jensen (1968),Henriksson and Merton (1981),Pesaran and
Timmermann (1994),Gruber (1996), and Carhart (1997) for “top-down” approaches that use return
series. See Jiang, Yao, and Yu (2007),Kaplan and Sensoy (2008), and Elton, Gruber, and Blake
(2011,2012) for “bottom-up” approaches that use changes in returns in response to changes in
portfolio holdings.
19 Odean (1999) weights trades equally, has different holding period lengths, and cannot risk-adjust
in his framework, whereas in counterfactual portfolios we retain the weighting of the original
trade and can risk-adjust.
18 U. Bhattacharya et al.
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
These counterfactual portfolios, most importantly, allow us to test whether portfolio
performance changes because an investor trades the “right” ETF at the “wrong” point in
time, or because an investor trades the “wrong” ETF at the “right” point in time, or both.
The wrong point in time is when an investor buys high and sells low. The right ETF is diffi-
cult to determine. As the ETF offerings are many, and are often based on narrow indices
(see Panels A, B, and C of Table II), a wrong ETF is one that promises suboptimal expected
Sharpe ratios. From an ex ante perspective, a single ETF (or alternatively, a basket of ETFs)
that replicates the market portfolio best would be the right ETF. Therefore, any ETF that
tracks only a tiny sub-market or has too large a tracking error with the market index is a
wrong ETF. Given these criteria, it seems that a sensible proxy for the right ETF in our con-
text is an ETF on the MSCI World Index.
20
We start our analysis with the non-ETF part of an investor’s portfolio. This is called
the “benchmark” portfolio (BM) because we want to see how the performance of an in-
vestor’s portfolio changes when ETFs are added to it. We then examine what would hap-
pen if ETFs were added to this benchmark portfolio, but were just bought-and-held. So
this first counterfactual portfolio is an investor’s full portfolio that includes all non-ETFs
and ETFs ever bought where we assume that the investor buys and holds but never sells
any ETF actually purchased.
21
This is called the buy-and-hold ETF portfolio (B&H). As
it is a buy-and-hold portfolio, it has no ETF timing. We then allow the ETFs to be traded,
which is the actual full portfolio (FULL) of the investor. By doing this, we introduce ETF
timing.
It is apparent from the above construction of the portfolios that the difference in returns
between the full portfolio, FULL, and the non-ETF part portfolio, BM, is the change in re-
turns from adding ETFs to an investor’s portfolio. So FULL minus BM is the ETFs impact
on portfolio performance. The important task is to decompose this differential return into
the contribution that is coming from ETF timing and the contribution that is coming from
ETF selection.
FULL minus BM, the ETF impact on portfolio performance, can be decomposed into
FULL minus B&H and B&H minus BM. As FULL is the actual full portfolio with actual
ETF buys and sells at different points in time, and B&H is the counterfactual full portfolio
20 Calvet, Campbell, and Sodini (2007) also use the MSCI World Index as the market benchmark for
Swedish investors holding portfolios containing stocks, funds, bonds, and other marketable secur-
ities. The MSCI World Index is a theoretically efficient choice from an ex ante perspective. This is
because the MSCI World Index, as a proxy of the market portfolio, promises the highest expected
Sharpe ratio, assuming that investors do not have private information and that capital markets are
semi-strong form efficient. We choose the Vanguard Global Stock Index Fund that tracks the
MSCI World Index. We choose this fund for many reasons. First, this fund is well known and
would have been available at an expense ratio of 0.5% p.a. to our investors throughout our entire
observation period. Alternatives like a value-weighted portfolio of all assets held by all investors
are not available as an ETF. Other well-known funds have inception dates that do not cover our
entire observation period. Note that all results hold qualitatively if we use an investable ETF on
the German DAX index instead of the MSCI World Index.
21 Results remain qualitatively unchanged if we create this first counterfactual portfolio using all
non-ETFs and only the first ETF ever bought. This means we disregard all purchases and sales of
ETFs after the initial purchase.
Abusing ETF 19
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
with a buy-and-hold strategy for ETFs that switches off timing, it is clear that FULL minus
B&H is the contribution that is coming from an investor’s ETF timing ability. So we meas-
ure an investor’s ETF timing ability using FULL minus B&H. B&H minus BM must then
be the contribution that is coming from ETF selection ability of the investor because it has
no timing in it. So we measure the ETF selection ability (relative to not choosing any ETFs)
of an investor using B&H minus BM.
The above decomposition, as it ignores what would happen if the investor who bought
a low-cost, well-diversified ETF—the strategy that is prescribed in classical finance
theory—is unable to analyze the opportunity loss by not doing so. It is for this reason that
we do an additional decomposition.
Using our full portfolio (FULL), we examine what would happen if we replace all ETF
trades with trades in a low-cost ETF on the MSCI World Index. This second counterfac-
tual portfolio is, therefore, the investor’s full portfolio where we replace all ETF buys and
sells with buy and sell trades in a Vanguard ETF that tracks the MSCI World Index. This
is called the Trade MSCI World (MSCI) portfolio. In this counterfactual portfolio, we get
rid of selection, because only investments in the ETF on the MSCI World Index are
included.
We then examine what would happen to the MSCI counterfactual portfolio if we allow
the low-cost ETF on the MSCI World Index to be just bought-and-held instead of being
traded. This third counterfactual portfolio is, therefore, an investor’s full portfolio where
we replace all ETF purchases of an investor with purchases of a Vanguard ETF that
tracks the MSCI World Index, and where we disregard all ETF sales, emulating a pure
Table VII. Our counterfactual portfolios
BM, “only non-ETF securities,” is the non-ETF portion of investors’ portfolios. This portfolio
serves as the benchmark in our analysis. ETF, “Only ETF securities,” is the ETF portion of in-
vestors’ portfolios. FULL is the actual full risky portfolio of investors. In B&H, “non-ETF
part þETF part with B&H,” we disregard all ETF sell transactions. In MSCI, “non-ETF part þETF
part with MSCI,” each ETF transaction is replaced by a transaction in a Vanguard fund that
tracks the MSCI World Index. In MBM, “non-ETF part þETF part with B&H MSCI,” we disregard
all ETF sell transactions and each ETF purchase is replaced by a purchase in a Vanguard fund
that tracks the MSCI World Index.
Portfolio Abbreviation Construction Description
Benchmark (non-
ETF part)
BM Only non-ETF securities The non-ETF part of the full
portfolio
ETF part ETF Only ETF securities The ETF part that of the full
portfolio
Full portfolio FULL Non-ETF part þETF part The actual full portfolio con-
sisting of the ETF part and
the non-ETF part
Full portfolio with
B&H ETF
B&H Non-ETF part þETF part
with B&H
The full portfolio with only
buy-and-hold ETFs
Full portfolio with
MSCI ETF
MSCI Non-ETF part þETF part
with MSCI
The full portfolio with all
ETFs replaced by MSCI
Market
benchmark
MBM Non-ETF part þETF part
with B&H MSCI
The full portfolio with buy-
and-hold MSCI
20 U. Bhattacharya et al.
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
buy-and-hold strategy of the market portfolio. This is called the “market benchmark”
(MBM) because we want to see what happens when an investor buys and holds the ETF on
the MSCI World Index (i.e., the “right” ETF). This counterfactual portfolio has neither
ETF timing nor selection in its ETF part. Table VII provides an overview of the counterfac-
tual portfolios.
It is apparent from our discussion of the hypothetical portfolios that FULL minus MBM
is the “opportunity loss” that the investor incurs by deviating his actual full portfolio from
a theoretically sound full portfolio. FULL minus MBM, the opportunity loss, can be decom-
posed into MSCI minus MBM and FULL minus MSCI. As the MSCI is the counterfactual
full portfolio with MSCI World Index buys and sells and the MBM is the counterfactual
portfolio with MSCI World Index buy-and-hold only, MSCI minus MBM is the change in
returns caused by trading instead of holding the MSCI World Index ETF. As the MSCI
World Index is our proxy for the market, this is the classical way to measure the impact of
market timing on portfolio performance, and following this tradition we call this measure
“market timing.” Note again that MSCI minus MBM is market timing (trading the MSCI
ETF minus buying-and-holding the MSCI ETF) and should not be confused with the previ-
ous ETF timing ability (FULL minus B&H, i.e., trading selected ETFs minus buying-and-
holding selected ETFs). As FULL is the actual full portfolio with actual ETF buys and sells
at different points in time, and MSCI is the counterfactual full portfolio with MSCI World
Index ETF buys and sells at the same points in time, it is clear that FULL minus MSCI is the
performance contribution that is coming from choosing these particular ETFs instead of the
MSCI World Index ETF. FULL minus MSCI, therefore, might be considered the classical
way (cf. e.g., Brinson, Hood, and Beebower, 1986) to measure the impact of security selec-
tion on portfolio performance. So we call this measure ETF selection ability (relative to
choosing MSCI).
We notice above that there are many ways of decomposing our returns. These could be
confusing. To simplify, though we show all our decompositions in Tables VIII–XI, we focus
on mostly interpreting two of our most important decompositions: FULL minus B&H,
which measures an investor’s ETF timing ability, and FULL minus MSCI, which measures
an investor’s ETF selection ability relative to passive indices. Note that FULL minus MSCI
measures the impact of choosing a particular ETF instead of the “right” MSCI ETF; it
should not be confused with the previous B&H minus BM ETF selection, which measures
ETF selection ability with respect to not choosing any ETFs.
We use several metrics to compare portfolio performance: the mean of the return, the
standard deviation of returns, the RSRL
22
(1 minus the quotient of the Sharpe ratio of an
investor’s portfolio over the Sharpe ratio of the MSCI World ACWI) over the sample period
in which ETFs are held, Jensen’s (1968) alpha, and the unsystematic variance share (aver-
age of the ratio of idiosyncratic variance divided by the total variance of the portfolio re-
turn). To compute Jensen’s alpha and the unsystematic variance share, we use the MSCI
ACWI as the benchmark. All measures are computed per investor and then averaged over
the cross-section of investors to make comparisons easier. Note that there are no qualitative
differences if we first average across investor per time period and then use the average over
the different time periods.
22 Calvet, Campbell, and Sodini (2007,2009) suggest using the RSRL to measure the under-
diversification in a household’s risky portfolio.
Abusing ETF 21
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
Table VIII. Counterfactual portfolios and performance decomposition
This table reports the results from a portfolio performance decomposition using counterfactual analysis at the investor level. If no ETF is held on day t, we re-
place the return with the return of the non-ETF part. Panel A displays results after transaction costs, whereas Panel B presents results before transaction costs.
Portfolios in the top half of each panel are created by either changing the securities considered (actual securities or Vanguard MSCI) or the trading strategy (ac-
tual behavior or B&H). The description of theses portfolios is in Table VII. The performance metrics are: return p.a. (average of time series mean returns of indi-
vidual investors), standard deviation p.a. (mean of the standard deviations of time series returns of individual investors), the RSRL, and measures based on a
one-factor model of performance evaluation using the MSCI ACWI as the benchmark index. Alpha represents Jensen’s alpha, p-value is a test of the alpha
against 0, beta is the coefficient on the market factor, and unsystematic variance share is the fraction of the variance the model is unable to explain. The column
“Return Difference from benchmark in %” is the difference in returns between the non-ETF part of the portfolio (BM) and the respective portfolio. The column
p-value of return difference from benchmark” presents the results of a test of whether the difference is statistically different from zero. The last column reports
the number of investors. The bottom half of each Panel in Table VIII decomposes the return contribution of ETFs. ETFs’ impact on portfolio performance is shown
in row 1 (calculated as: FULL minus BM). The decomposition of row 1 into ETF timing ability (FULL minus B&H) and ETF selection ability (relative to not choosing
ETFs) (B&H minus BM) is shown in rows 2 and 3, respectively. The opportunity loss they have incurred by not investing into a theoretically sound ETF is shown
in row 4 (FULL minus MBM). The decomposition of row 4 into market timing (MSCI minus MBM) and ETF selection (relative to choosing MSCI) (FULL minus
MSCI) is shown in rows 5 and 6, respectively. The columns p-value of return (RSRL, alpha) difference” present p-values of a test of whether the return (RSRL,
alpha) difference is statistically different from zero. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.
Panel A: Net of transaction costs
Portfolio Description Return
p.a. in %
Standard
deviation
p.a. in %
Relative Sharpe
Ratio Loss in %
Alpha
p.a. in %
p-Value
of Alpha
against 0
Beta Unsystematic
variance
share in %
Return difference
from Benchmark
in %
p-Value of return
difference from
Benchmark
N
BM Non-ETF part 3.91 23.48 45.49 0.24 0.710 0.75 39.41 0.00 n.a. 1,061
ETF ETF Part 0.55 24.37 85.40 2.72 0.000*** 0.75 39.52 4.46 0.000*** 1,061
FULL Non-ETF part þETF part 2.74 21.96 54.64 0.35 0.53 0.74 34.54 1.16 0.058*** 1,061
B&H Non-ETF part þETF part
with B&H
3.52 20.87 51.82 0.76 0.054* 0.73 30.98 0.39 0.547 1,061
MSCI Non-ETF part þETF part
with MSCI
4.02 22.09 37.73 0.77 0.190 0.75 33.66 0.12 0.913 1,061
MBM Non-ETF part þETF part
with B&H MSCI
4.43 20.36 36.63 1.43 0.000*** 0.73 30.03 0.52 0.630 1,061
(continued)
22 U. Bhattacharya et al.
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
Table VIII. Continued
Difference Performance decomposition Return
difference
p-Value of
return
difference
RSRL
difference
p-Value of
RSRL
difference
Alpha
difference
p-Value of
alpha
difference
FULL BM ETF’s impact on portfolio performance 1.16 0.058*** 9.15 0.000*** 0.11 0.687
FULL B&H ETF timing ability 0.77 0.075* 2.82 0.071* 1.11 0.001***
B&H BM ETF selection ability (relative to not choosing ETFs) 0.39 0.547 6.33 0.006*** 1.00 0.018**
FULL MBM Opportunity loss 1.69 0.007*** 18.01 0.000*** 1.79 0.000***
MSCI MBM Market timing 0.41 0.236 1.10 0.436 0.67 0.074*
FULL MSCI ETF selection ability (relative to choosing MSCI) 1.28 0.022** 16.91 0.000*** 1.12 0.000***
Panel B: Gross of transaction costs
Portfolio Description Return
p.a. in %
Standard
deviation
p.a. in %
Relative
Sharpe Ratio
Loss in %
Alpha
p.a. in %
p-Value of
Alpha against 0
Beta Unsystematic
variance
share in %
Return difference
from Benchmark
in %
p-Value of return
difference from
Benchmark
N
BM Non-ETF part 5.51 23.24 35.72 1.18 0.041** 0.75 39.18 0.00 n.a. 1,061
ETF ETF Part 1.02 24.20 74.57 1.24 0.013** 0.75 39.25 4.49 0.000*** 1,061
FULL Non-ETF part þETF part 4.18 21.78 45.15 0.97 0.043** 0.74 34.27 1.33 0.033** 1,061
B&H Non-ETF part þETF part
with B&H
4.75 21.56 43.36 1.56 0.000*** 0.75 31.57 0.76 0.234 1,061
MSCI Non-ETF part þETF part
with MSCI
5.47 21.92 28.53 2.09 0.000*** 0.75 33.34 0.04 0.970 1,061
MBM Non-ETF part þETF part
with B&H MSCI
5.41 20.28 28.73 2.26 0.000*** 0.73 29.67 0.10 0.927 1,061
(continued)
Abusing ETF 23
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
Table VIII. Continued
Difference Performance decomposition Return
difference
p-Value of
return
difference
RSRL
difference
p-Value of
RSRL
difference
Alpha
difference
p-Value of
alpha
difference
FULL BM ETF’s impact on portfolio performance 1.33 0.033*** 9.43 0.000*** 0.22 0.415
FULL B&H ETF timing ability 0.57 0.215 1.80 0.354 0.59 0.038**
B&H BM ETF selection ability (relative to not choosing ETFs) 0.76 0.234 7.63 0.000*** 0.37 0.265
FULL MBM Opportunity loss 1.23 0.045** 16.43 0.000*** 1.29 0.000***
MSCI MBM Market timing 0.06 0.839 0.20 0.884 0.17 0.542
FULL MSCI ETF selection ability (relative to choosing MSCI) 1.29 0.019** 16.62 0.000*** 1.12 0.000***
24 U. Bhattacharya et al.
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
Table IX. Trading behavior and investment performance across turnover quintiles
In this table, all users of ETFs are divided into five quintiles in terms of their average monthly portfolio turnover before they start using ETFs. Quintile 1 has the
lowest turnover and quintile 5 has the highest turnover. In Panel A, the mean turnover of each quintile group before ETF use is shown in row 1, portfolio turnover
after ETF use is shown in row 2, ETF turnover after ETF use is shown in row 3, and non-ETF turnover after ETF use is shown in row 4. In Panel B, the ETFs’ impact
on portfolio performance (net of transactions costs) is shown in row 1 (calculated as: FULL minus BM of top half in Table VIII). The decomposition of row 1 into
ETF timing ability (net) (FULL minus B&H in top half of Table VIII) and ETF selection ability (relative to not choosing ETFs) (net) (B&H minus BM in top half of
Table VIII) is shown in rows 2 and 3, respectively. The opportunity loss (net) they have incurred by not investing into a theoretically sound ETF is shown in row 4
(FULL minus MBM in top half of Table VIII). The decomposition of row 4 into market timing (net) (MSCI minus MBM in top half of Table VIII) and ETF selection
(relative to choosing MSCI) (net) (FULL minus MSCI in top half of Table VIII) is shown in rows 5 and 6, respectively. ETFs’ impact on portfolio diversification (net
RSRL) is shown in row 7 (net RSRL of FULL minus net RSRL of BM in top half of Table VIII) and opportunity loss from portfolio diversification (net RSRL) is shown
in row 8 (net RSRL of portfolio FULL minus net RSRL of portfolio MBM in top half of Table VIII). Sample size, 902, is different from previous tables because we
only include users of ETFs who do not hold ETFs throughout our sample period. The column p-value” presents p-values of a test of whether the difference be-
tween Q1 and Q5 is different from zero on a per investor level. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.
Q1 (lowest) Q2 Q3 Q4 Q5 (highest) Difference t-test
Number of investors 181 180 181 180 180 Q1Q5 p-value
Panel A: Trading behavior
Portfolio turnover before ETF use (% per month) 1.36 3.16 5.56 9.80 30.30 28.94 0.000***
Portfolio turnover after ETF use (% per month) 1.84 2.18 5.09 7.34 16.20 14.35 0.000***
ETF turnover after ETF use (% per month) 2.80 4.60 5.98 7.95 15.94 13.14 0.000***
Non-ETF turnover after ETF use (% per month) 1.71 1.69 4.33 6.35 15.67 13.97 0.000***
Panel B: Portfolio performance/portfolio diversification
ETFs’ impact on portfolio performance FULL BM 0.64** 2.09*** 0.51 0.96** 2.98 2.35 0.492
ETF timing ability FULL B&H 0.08 0.29 1.11** 1.82 3.81*** 3.74 0.009***
ETF selection ability (relative to not choosing ETFs) B&H BM 0.56 2.38*** 0.60 2.78 0.83 1.39 0.626
Opportunity loss FULL MBM 0.90*** 1.93*** 1.29*** 2.82* 1.19 2.09 0.583
Market timing MSCI MBM 0.65** 0.84*** 0.57 1.07* 3.87** 4.52 0.004***
ETF selection ability (relative to choosing MSCI) FULL MSCI 1.62*** 2.49*** 1.83*** 2.06*** 1.24 2.86 0.357
ETFs’ impact on portfolio diversification (net RSRL) FULL BM 0.05 0.13*** 0.09** 0.19*** 0.11** 0.05 0.361
Opportunity loss from portfolio diversification (net RSRL) FULL MBM 0.17* 0.11*** 0.16*** 0.25*** 0.36*** 0.19 0.157
Abusing ETF 25
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
Table X. Trading behavior and investment performance across portfolio value quintiles
In this table, all investors (users of ETFs and non-users of ETFs) divided into five quintiles in terms of their average monthly portfolio value before they start using
ETFs. Quintile 1 has the lowest value and quintile 5 has the highest value. In Panel A, the mean turnover of each quintile group before ETF use is shown in row 1,
portfolio turnover after ETF use is shown in row 2, ETF turnover after ETF use is shown in row 3, and non-ETF turnover after ETF use is shown in row 4. In Panel
B (Portfolio Performance/Portfolio Diversification), the ETFs’ impact on portfolio performance (net of transactions costs) is shown in row 1 (calculated as: FULL
minus BM of top half in Table VIII). The decomposition of row 1 into ETF timing ability (net) (FULL minus B&H in top half of Table VIII) and ETF selection ability
(relative to not choosing ETFs) (net) (B&H minus BM in top half of Table VIII) is shown in rows 2 and 3, respectively. The opportunity loss (net) they have incurred
by not investing into a theoretically sound ETF is shown in row 4 (FULL minus MBM in top half of Table VIII). The decomposition of row 4 into market timing (net)
(MSCI minus MBM in top half of Table VIII) and ETF selection (relative to choosing MSCI) (net) (FULL minus MSCI in top half of Table VIII) is shown in rows 5 and
6, respectively. ETFs’ impact on portfolio diversification (net RSRL) is shown in row 7 (net RSRL of FULL minus net RSRL of BM in top half of Table VIII) and op-
portunity loss from portfolio diversification (net RSRL) is shown in row 8 (net RSRL of portfolio FULL minus net RSRL of portfolio MBM in top half of Table VIII).
Sample size, 902, is different from previous tables because we only include users of ETFs who do not hold ETFs throughout our observation period. The column
p-value” presents p-values of a test of whether the difference between Q1 and Q5 is different from zero on a per investor level. ***, **, and * denote statistical
significance at the 1%, 5%, and 10% levels, respectively.
Q1 (lowest) Q2 Q3 Q4 Q5 (highest) Difference t-Test
Number of investers 181 180 181 180 180 Q1Q5 p-Value
Panel A: Trading behavior
Portfolio turnover before ETF use (% per month) 13.78 11.85 8.74 8.56 7.16 6.62 0.000***
Portfolio turnover after ETF use (% per month) 7.16 7.85 5.05 6.10 6.46 0.70 0.587
ETF turnover after ETF use (% per month) 6.97 7.58 7.10 7.42 8.19 1.23 0.448
Non-ETF turnover after ETF use (% per month) 6.66 6.98 5.02 5.33 5.72 0.94 0.397
Panel B: Portfolio performance/portfolio diversification
ETFs’ impact on portfolio performance FULL BM 4.12 0.79 1.53*** 0.15 0.87 3.25 0.333
ETF timing ability FULL B&H 1.09 1.55*** 0.87 0.84 0.22 0.86 0.347
ETF selection ability (relative to not choosing ETFs) B&H BM 3.03 0.75 0.65 0.69 0.65 2.38 0.373
Opportunity loss FULL MBM 1.07 0.94 2.23 1.89 1.76 1.15 0.447
Market timing MSCI MBM 0.24 0.54 0.48 1.07 0.02 0.37 0.747
ETF selection ability (relative to choosing MSCI) FULL MSCI 0.26 1.95*** 2.62*** 0.02 1.95*** 1.70 0.571
ETFs’ impact on portfolio diversification (net RSRL) FULL BM 0.15*** 0.09** 0.11*** 0.06* 0.17** 0.02 0.799
Opportunity loss from portfolio diversification (net RSRL) FULL MBM 0.39*** 0.20** 0.15*** 0.06 0.25*** 0.14 0.305
26 U. Bhattacharya et al.
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
Table XI. Trading behavior and investment performance across RSRL quintiles
In this table, all investors (users of ETFs and non-users of ETFs) are divided into five quintiles in terms of their average monthly RSRL before they start using
ETFs. Quintile 1 has the lowest RSRL while quintile 5 has the highest RSRL. In Panel A, the mean turnover of each quintile group before ETF use is shown in row
1, portfolio turnover after ETF use is shown in row 2, ETF turnover after ETF use is shown in row 3, and non-ETF turnover after ETF use is shown in row 4. In
Panel B, the ETFs’ impact on portfolio performance (net of transactions costs) is shown in row 1 (calculated as: FULL minus BM of top half in Table VIII). The de-
composition of row 1 into ETF timing ability (net) (FULL minus B&H in top half of Table VIII) and ETF selection ability (relative to not choosing ETFs) (net) (B&H
minus BM in top half of Table VIII) is shown in rows 2 and 3, respectively. The opportunity loss (net) they have incurred by not investing into a theoretically sound
ETF is shown in row 4 (FULL minus MBM in top half of Table VIII). The decomposition of row 4 into market timing (net) (MSCI minus MBM in top half of Table
VIII) and ETF selection (relative to choosing MSCI) (net) (FULL minus MSCI in top half of Table VIII) is shown in rows 5 and 6, respectively. ETFs’ impact on port-
folio diversification (net RSRL) is shown in row 7 (net RSRL of FULL minus net RSRL of BM in top half of Table VIII) and opportunity loss from portfolio diversifica-
tion (net RSRL) is shown in row 8 (net RSRL of portfolio FULL minus net RSRL of portfolio MBM in top half of Table VIII). Sample size, 902, is different from
previous tables because we only include users of ETFs who do not hold ETFs throughout our sample period. The column p-value” presents p-values of a test of
whether the difference between Q1 and Q5 is different from zero on a per investor level. ***, **, and * denote statistical significance at the 1%, 5%, and 10% lev-
els, respectively.
Q1 (lowest) Q2 Q3 Q4 Q5 (highest) Difference t-test
Number of investors 181 180 181 180 180 Q1 Q5 p-value
Panel A: Trading behavior
Portfolio turnover before ETF use (% per month) 9.54 7.82 9.90 11.38 11.47 1.92 0.166
Portfolio turnover after ETF use (% per month) 6.79 4.84 7.08 7.08 6.82 0.04 0.972
ETF turnover after ETF use (% per month) 6.74 6.21 7.15 9.33 7.82 1.08 0.426
Non-ETF turnover after ETF use (% per month) 6.20 4.10 7.47 5.70 6.24 0.04 0.969
Panel B: Portfolio performance/portfolio diversification
ETFs’ impact on portfolio performance FULL BM 0.52 0.71 1.55*** 3.83 0.56 0.04 0.972
ETF timing ability FULL B&H 1.09 1.76* 0.70 1.26 0.26 1.35 0.488
ETF selection ability (relative to not choosing ETFs) B&H BM 1.62 1.06 0.85 2.57 0.30 1.32 0.566
Opportunity loss FULL MBM 2.95 1.57 2.26 2.68 1.62 1.15 0.449
Market timing MSCI MBM 0.04 1.83 0.79 0.56 0.22 0.37 0.761
ETF selection ability (relative to choosing MSCI) FULL MSCI 1.82*** 1.51 2.18*** 0.86 2.10*** 0.28 0.763
ETFs’ impact on portfolio diversification (net RSRL) FULL BM 0.13*** 0.03 0.16*** 0.08** 0.17** 0.04 0.679
Opportunity loss from portfolio diversification (net RSRL) FULL MBM 0.25*** 0.16*** 0.25*** 0.10*** 0.29** 0.05 0.751
Abusing ETF 27
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
A question that naturally arises is how to compute returns for the ETF part of the port-
folio during months in which a previous ETF user does not hold any ETFs. As we construct
the counterfactual portfolios based on the non-ETF part of the portfolio plus different strat-
egies in the ETF part, the ETF share in the counterfactual portfolios is zero in these months,
while the return on the total counterfactual portfolio is equal to the return of the non-ETF
part. This is equivalent to the assumption that when the investor sells ETFs, the cash goes
toward purchasing non-ETF risky securities. We use this option in all our tables with one
notable exception: computing the return on the ETF part of the portfolio. When we analyze
only the ETF part of the portfolio, another option becomes viable. We could assume that
when the investor bought (sold) ETFs, the cash came from (went to) the cash account. So
we should use zero as return, instead of the non-ETF return, for the ETF part of the port-
folio in months without ETF holdings. The two options give us different results for the ETF
part of the portfolio. We discuss these differences later.
5.2 Results from Counterfactual Analysis
Table VIII shows the results of our portfolio return decomposition. Panel A shows the re-
sults net of transaction costs and Panel B shows the results for gross returns. To compute
the net returns for the counterfactual portfolios that use the Vanguard MSCI World ETF,
we retain the costs of the original transactions. The transaction costs on the Vanguard
MSCI World ETF are likely to be lower.
23
Panel A of Table VIII, which displays net returns, shows that an ETF investor decreases
the net return of his portfolio from 3.91% (BM) to 2.74% (FULL) per year. This drop of
1.16% (FULL minus BM)
24
is statistically significant at the 10% level. However, if this in-
vestor had just bought and held ETFs instead of trading in them, the increase in return would
be 0.77%. So the investor’s ETF timing ability is 0.77% (FULL minus B&H), which is stat-
istically significant at the 10% level. As the drop of 0.77% is the only part that is significant
in the overall drop of 1.16%, we conclude that it is mainly the investor’s poor ETF timing
ability that is adversely affecting portfolio’s return. If we adjust for risk and consider alphas,
we find that it is also the investor’s poor ETF timing ability that is adversely affecting the
risk-adjusted return of the portfolio [ETF timing ability expressed in alpha is 1.11% (FULL
minus B&H), which is statistically significant]. If we consider diversification and look at
RSRL, we find that most of the diversification loss of 9.15% is again coming from poor ETF
selection (6.33%) and both are statistically significant at 1% level.
We also examine what would happen if the investor chooses a MSCI World Index ETF.
The results are given at the bottom of Panel A of Table VIII. An investor would have im-
proved portfolio returns if all ETF trades had been replaced by a buy-and-hold strategy in a
low-cost, diversified ETF like the Vanguard MSCI World ETF. The opportunity loss for
doing this is a statistically significant 1.69% (FULL minus MBM). Most of this opportun-
ity loss, 1.28% (FULL minus MSCI), comes from ETF selection (relative to choosing
MSCI), and this number is statistically significant at the 5% level. If we adjust for risk and
examine the alphas, we find an opportunity loss of 1.79% (FULL minus MBM). Most of
this opportunity loss, 1.12% (FULL minus MSCI), again comes from ETF selection (rela-
tive to choosing MSCI) and this number is statistically significant at the 1% level. If we con-
sider portfolio diversification and look at RSRL, we also find that most of the 18.01%
23 As all our qualitative results hold also for gross returns, this assumption is not critical.
24 FULL minus BM has a rounding error.
28 U. Bhattacharya et al.
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
diversification loss is coming from ETF selection (relative to choosing MSCI) (16.91%),
and both are statistically significant at the 1% level.
Panel B of Table VIII gives the results of the counterfactual analysis before transaction
costs (i.e., gross returns). An investor who uses ETFs decreases their gross portfolio returns
from 5.51% (BM) to 4.18% (FULL). This decline of 1.33% (FULL minus BM) is statistic-
ally significant. However, if decomposed, none of the decomposed parts are statistically sig-
nificant. If we adjust for risk and consider the alphas, poor ETF timing seems to be
responsible for the decline.
When we examine opportunity loss at the bottom of Panel B in Table VIII, we find that
an investor would have higher portfolio returns if they had employed a buy-and-hold strat-
egy in a low-cost, diversified ETF like the Vanguard MSCI World ETF instead of conduct-
ing ETF trades. The opportunity loss they incur by trading ETFs in terms of gross returns is
a statistically significant 1.23% (FULL minus MBM). Most of this opportunity loss,
1.29% (FULL minus MSCI), comes from ETF selection (relative to choosing MSCI) and
is statistically significant. If we adjust for risk and examine the alphas, we also find an op-
portunity loss of 1.29% (FULL minus MBM). Most of this opportunity loss, 1.12%
(FULL minus MSCI), again comes from poor ETF selection (relative to choosing MSCI)
and is statistically significant. If we consider portfolio diversification and look at RSRL, we
find that most of the 16.43% diversification loss is coming from poor ETF selection (rela-
tive to choosing MSCI) (16.62%) and both are statistically significant.
To summarize, there are four major findings from Table VIII. First, an investor is hurt-
ing his overall portfolio performance mostly by poor ETF timing. Second, the cost from
poor market timing cannot all be due to extra costs from excessive ETF trading because we
see the same result for gross returns. Third, if an investor had added a MSCI World Index
ETF to their portfolios and applied a buy-and-hold strategy, their net and gross returns
would have improved significantly. Most of that improvement would have come from
replacing the ETFs actually traded by the MSCI World Index ETF (security selection), ra-
ther than by implementing a buy-and-hold strategy in the MSCI World Index ETF (market
timing). Fourth, the results for risk-adjusted returns (alphas) and portfolio diversification
(RSRL) are qualitatively similar to those for raw returns. Note that despite the fact that
ETFs make up only a fraction of 15% of the average investor’s full portfolio (Figure 1), we
do find that the investor’s unwise use adversely and significantly affects the return of the
full portfolio most of the time.
5.3 Discussion of Alternative Explanations
The results in Table VIII allow us to rule out three alternative hypotheses as explanations
for our results. The first alternative hypothesis is that the non-ETF part of an ETF investor’s
portfolio is responsible for the lack of improvement of portfolio performance after ETF use
rather than the ETF part. The net return on the ETF part is 0.55% if we assume that pur-
chases (and sales) of ETFs are financed from the non-ETF part, as is the maintained as-
sumption in this table. This number changes to 1.46% (unreported result) if we assume
that purchases (and sales) of ETFs are financed from (finance) the cash account, the second
option we discuss above. Whether it is 0.55% or 1.46%, the net return on the ETF part
of the portfolio is lower than the net return on the non-ETF part (3.91%). Further tests
show that they are statistically significantly lower. This rules out that the non-ETF part is
responsible for the lack of improvement in portfolio performance after ETF use.
Abusing ETF 29
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
The second alternative hypothesis is that investors sacrifice returns by using ETFs as
hedges and benefit from substantial diversification effects. If ETFs were used for hedging,
the RSRL of the full portfolio should be smaller for the FULL portfolio than for the BM
portfolio. Note that the net RSRL is higher for the FULL portfolio (54.64) than it is for the
BM benchmark non-ETF portfolio (45.49).
The third alternative hypothesis is that investors simply trade more in ETFs than in
other securities and that extra trading costs are the main cause of deterioration in net
returns. If we assume that buys (and sells) of ETFs are financed from the non-ETF por-
tion of the portfolio, as is the assumption for Table VIII, the return net of transaction
costs is –0.55% for the ETF part of the portfolio (ETF) and 3.91% for the non-ETF
part of the portfolio (BM). The return before transaction costs is 1.02% for the ETF
part of the portfolio (ETF) and 5.51% (BM) for the non-ETF part of the portfolio.
Although the drop in returns due to transactions is marginally larger for the ETF part
than for the non-ETF part, the important insight is that the performance difference al-
ready exists for gross returns. Therefore, the costs that investors incur in trading ETFs
are not the only reason why the ETF part of the portfolio under-performs the non-ETF
part.
25
5.4 Impact of Investor Heterogeneity
In this section, we explore the role of investor heterogeneity in the use of ETFs. Specifically,
we examine whether overconfident investors and/or financially unsophisticated investors
trade ETFs unwisely. The research design is to check whether there is a difference in trading
behavior and portfolio performance among ETF users sorted into quintiles along these two
dimensions.
Overconfident investors have higher portfolio turnover (e.g., Barber and Odean, 2000),
so we use turnover as our first sorting variable. Financially unsophisticated investors have
lower portfolio values, and are typically under-diversified (e.g., Goetzmann and Kumar,
2008). We measure under-diversification using the RSRL. We use portfolio value and port-
folio diversification as our second and third sorting variables.
We use several metrics as measures of the trading behavior of ETF users. We first use
the portfolio turnover in ETF users’ portfolios before and after using ETFs. Then we de-
compose the turnover after ETF use into turnover in the ETF part and turnover in the non-
ETF part of the portfolio. To measure the impact of portfolio performance, we use the
same return differentials as in Table VIII, but we focus our attention to ETF timing ability
(FULL minus B&H) and ETF selection (relative to choosing MSCI) (FULL minus MSCI).
To check for a portfolio diversification impact, we again resort to the change in RSRL.
In Table IX, ETF users are grouped into quintiles depending on their average portfolio
turnover before they start using ETFs. Quintile 1 has the lowest turnover whereas quintile 5
has the highest.
ETF users, who trade more than other users before ETF use, also trade more than their
peers after ETF use. This holds for both the ETF part and for the non-ETF part of their
25 We obtain the same conclusion if we assume that purchases (and sales) of ETFs are financed
from (finance) the cash part of the portfolio (results not tabulated). Then the after transaction
costs return is 1.46% for the ETF part of the portfolio and 2.10% for the non-ETF part of the
portfolio; the return before transaction costs is 0.715% for the ETF part and 0.722% for the
non-ETF part.
30 U. Bhattacharya et al.
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
portfolio. Within each turnover quintile, ETF turnover is higher than the non-ETF turnover
after ETF use. Taken together, this suggests that the availability of ETFs induces the active
traders to remain active, but to shift some of their active trading from non-ETFs to ETFs.
Are there differences in performance, timing, and selection abilities, or portfolio diversi-
fication between the investor quintiles? Table IX shows some systematic relations.
First, no quintiles have statistically significant gains by trading ETFs (FULL minus BM).
However, overconfident investors, as measured by high portfolio turnover, have much
worse ETF timing abilities (FULL minus B&H), but actually have better ETF selection abil-
ities. Third, in terms of opportunity loss, almost all quintiles significantly lose by not buying
and holding the MSCI World Index ETF. However, turnover does not seem to be related to
ETF selection. Fourth, almost all quintiles worsen their portfolio diversification as meas-
ured by RSRL.
Table X is analogous to Table IX except that ETF users are grouped into five quintiles
according to their average portfolio value before they start using ETFs. Quintile 1 has the
lowest portfolio value whereas quintile 5 has the highest. We find a negative relation be-
tween investor sophistication as measured by portfolio value and trading before ETF use,
but not after ETF use. As in Table IX the level of turnover across all quintiles after ETF use
is higher for the ETF portion of the portfolio than for the non-ETF part.
We next examine whether wealth differences between users, as measured by portfolio
value, affect portfolio performance, timing, and selection abilities, and portfolio diversifica-
tion. The results in Table X indicate no systematic relation. Nevertheless, we again find
that there is no distinct investor group that significantly benefits from ETF use or that ex-
periences significant increases in diversification (as measured by RSRL).
Table XI is analogous to Table X except that ETF users are grouped into five quintiles
depending on their RSRL before they start using ETFs. Quintile 1 has the lowest RSRL
(highest portfolio diversification), whereas quintile 5 has the highest RSRL (lowest port-
folio diversification). We find that with increasing RSRL, the portfolio turnover of ETF
users’ increases. This positive relation exists before and after ETF use and is driven by trad-
ing in the non-ETF part of the portfolio. Again, as in Table IX the level of portfolio turn-
over across all quintiles after ETF use is much higher for the ETF portion of the portfolio
than for the non-ETF part.
We next examine whether RSRL differences are related to performance, timing, and se-
lection abilities. The results in Table XI indicate no systematic relation. Although we find
that there is no quintile in which investors benefit from ETF use, and almost all quintiles
lose (sometimes significantly) by not buying and holding the MSCI World Index ETF, we
do not see significant cross-sectional differences across the quintiles in terms of perform-
ance, timing, and selection abilities.
To summarize our exploration of investor heterogeneity, there is no distinct group of in-
vestors whose portfolio performance is positively affected by the use of ETFs, no matter
which measure (performance, timing, selection, or RSRL) or sort (turnover, portfolio value,
or RSRL) we examine. We also find that no groups will lose by investing in the right MSCI
ETF. Our sorting exercise also yields one potential explanation. Investors from virtually all
groups do not substantially adapt their trading behavior after ETF use. Those who traded
more before ETF use continue to trade more after ETF use, both in the ETF portion of the
portfolio, as well as in the non-ETF part. Investors therefore appear to make the same mis-
takes when they trade ETFs that they have made in trading non-ETFs.
Abusing ETF 31
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
6. Conclusion
In this paper, we investigate whether a sample of individual investors in Germany benefit
from using ETFs in the period 2005–2010. We find that the portfolio performance of ETF
users relative to non-users does not improve after ETF use. This is primarily due to buying
ETFs at the “wrong” time. There is also an opportunity loss that results mostly from not
choosing ETFs that are low-cost and well-diversified. Therefore, for the individual investors
in our sample, buying and holding well-diversified, low-cost ETFs would have been a wise
strategy. This strategy, of course, also saves transaction costs.
The innovation of passive ETFs, with its enormous potential to act as a low-cost and li-
quid investment vehicle for diversification, may not help individual investors to enhance
their portfolio performance. Problems arise when they actively abuse passive ETFs by buy-
ing and selling them at the “wrong” time or trading the “wrong” ETFs (buying and selling
ETFs that are linked to narrow indices). Ironically, the growth in the number of ETFs that
track single industries or countries seems to encourage this damaging behavior.
Our findings will make policymakers, regulators, consumer protection agencies, compa-
nies with 401(k) plans, financial planners, and financial economists more cautious when
recommending ETF use. From a policy perspective, therefore, programs promoting savings
in well-diversified, low-cost ETFs that simultaneously limit the potential to actively trade in
them might be beneficial to individual investors.
References
Ahearne, A. G., Griever, W. L., and Warnock, F. E. (2004) Information costs and home bias: an
analysis of US holdings of foreign securities, Journal of International Economics 62, 313–336.
Barber, B. M. and Odean, T. (2000) Trading is hazardous to your wealth: the common stock in-
vestment performance of individual investors, The Journal of Finance 55, 773–806.
Barber, B. M. and Odean, T. (2002) Online investors: do the slow die first?, Review of Financial
Studies 15, 455–487.
Bhattacharya, U., Hackethal, A., Kaesler, S., Loos, B., and Meyer, S. (2012) Is unbiased financial
advice to retail investors sufficient? Answers from a large field study, Review of Financial
Studies 25, 975–1032.
Blume, M. and Friend, I. (1975) The allocation of wealth to risky assets—the asset structure of in-
dividual portfolios and some implications for utility functions, The Journal of Finance 30,
585–603.
Boldin, M. D. and Cici, G. (2010) The index fund rationality paradox, Journal of Banking &
Finance 34, 33–43.
Brinson, G. P., Hood, L. R., and Beebower, G. L. (1986) Determinants of portfolio performance,
Financial Analysts Journal 51, 133–138.
Calvet, L. E., Campbell, J. Y., and Sodini, P. (2007) Down or out: assessing the welfare costs of
household investment mistakes, Journal of Political Economy 115, 707–747.
Calvet, L. E., Campbell, J. Y., and Sodini, P. (2009) Measuring the financial sophistication of
households, American Economic Review 99, 393–398.
Carhart, M. M. (1997) On persistence in mutual fund performance, The Journal of Finance 52,
57–82.
Choi, J. J., Laibson, D., and Madrian, B. C. (2010) Why does the law of one price fail? An experi-
ment on index mutual funds, Review of Financial Studies 23, 1405–1432.
Cooper, I. and Kaplanis, E. (1994) Home bias in equity portfolios, inflation hedging and interna-
tional capital market equilibrium, Review of Financial Studies 7, 45–60.
32 U. Bhattacharya et al.
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
Deutsche Bo¨ rse (2010) Facts & Figures 10 Jahre ETF-Handel auf Xetra in 2013, Deutsche Bo¨ rse,
Frankfurt.
Dickerson, M. D. and Gentry, J. W. (1983) Characteristics of adopters and non-adopters of home
computers, Journal of Consumer Research 10, 225–235.
Elton, E. J., Gruber, M. J., and Blake, C. R. (2011) Holdings data, security returns and the selec-
tion of superior mutual funds, Journal of Financial and Quantitative Analysis 46, 341–367.
Elton, E. J., Gruber, M. J., and Blake, C. R. (2012) An examination of mutual fund timing ability
using monthly holdings data, Review of Finance 16, 619–645.
Elton, E. J., Gruber, M. J., and Busse, J. A. (2004) Are investors rational? Choices among index
funds, The Journal of Finance 59, 261–288.
Fama, E. F. and French, K. R. (1993) Common risk factors in the returns on stocks and bonds,
Journal of Financial Economics 33, 3–56.
Frame, S. W. and White, L. J. (2004) Empirical studies of financial innovation: lots of talk, little
action?, Journal of Economic Literature 42, 116–144.
French, K. R. (2008) Presidential address: the cost of active investing, The Journal of Finance 63,
1537–1573.
French, K. R. and Poterba, J. (1991) Investor diversification and international equity markets,
American Economic Review 81, 222–226.
Goetzmann, W. N. and Kumar, A. (2008) Equity portfolio diversification, Review of Finance 12,
433–463.
Gruber, M. J. (1996) Another puzzle: the growth in actively managed mutual funds, The Journal
of Finance 51, 783–810.
Henriksson, R. D. and Merton, R. C. (1981) On market timing and investment performance II.
Statistical procedures for evaluating forecasting skills, Journal of Business 54, 513–533.
Hortac¸su, A. and Syverson, C. (2004) Product differentiation, search costs, and competition in the
mutual fund industry: a case study of S&P 500 index funds, The Quarterly Journal of
Economics 119, 403–456.
Huberman, G. (2001) Familiarity breeds investment, Review of Financial Studies 14, 659–680.
Jensen, M. C. (1968) The performance of mutual funds in the period 1945–1964, The Journal of
Finance 23, 389–416.
Jiang, G. J., Yao, T., and Yu, T. (2007) Do mutual funds time the market? Evidence from portfolio
holdings, Journal of Financial Economics 86, 724–758.
Kaplan, S. N. and Sensoy, B. A. (2008) Do mutual funds time their benchmarks?, Mimeo.
Kelly, M. (1995) All their eggs in one basket: portfolio diversification of U.S. households, Journal
of Economic Behavior and Organization 27, 87–96.
Lewis, K. K. (1999) Trying to explain home bias in equities and consumption, Journal of
Economic Literature 37, 571–608.
Markowitz, H. M. (1952) Portfolio selection, The Journal of Finance 7, 77–91.
Mu¨ ller, S. and Weber, M. (2010) Financial literacy and mutual fund investments: who buys ac-
tively managed funds?, Schmalenbach Business Review 62, 126–153.
Odean, T. (1999) Do investors trade too much?, American Economic Review 89, 1278–1298.
Pesaran, M. H. and Timmermann, A. G. (1994) A generalization of the non-parametric
Henriksson–Merton test of market timing, Economics Letters 44, 1–7.
Petersen, M. A. (2009) Estimating standard errors in finance panel data sets: comparing
approaches, Review of Financial Studies 22, 435–480.
Poterba, J. M. and Shoven, J. B. (2002) Exchange-traded funds: a new investment option for tax-
able investors, American Economic Review 92, 422–427.
Ramaswamy, S. (2011) Market structures and systemic risks of exchange-traded funds, Mimeo.
Seasholes, M. and Zhu, N. (2010) Individual investors and local bias, The Journal of Finance 65,
1987–2010.
Abusing ETF 33
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
Sharpe, W. (1964) Capital asset prices: a theory of market equilibrium under conditions of risk,
The Journal of Finance 19, 425–442.
Tobin, J. (1958) Liquidity preference as behavior towards risk, Review of Economic Studies 25,
65–86.
Treynor, J. L. and Mazuy, K. K. (1966) Can mutual funds outguess the market?, Harvard Business
Review 44, 131–163.
World Federation of Exchanges. (2013) Statistics—Annual Query Tool. Retrieved from: http://
www.world-exchanges.org.
Zhu, N. (2002) The local bias of individual investors, Mimeo.
34 U. Bhattacharya et al.
at University Mannheim on August 17, 2016http://rof.oxfordjournals.org/Downloaded from
... Ben-David et al. [16] Tse [125] Ben-David et al. [18] Bhattacharya et al. [19] Broman [23] Sherrill et al. [115] Lettau and Madhavan [80] Caginalp et al. [26] Brown et al. [25] Hsu et al. [59] Ackert and Tian [2] Fiordelisi [46] Broman and Shum [24] Kwon [74] Bhojraj et al. [20] Pornpikul and Nettayanun [105] Investment Company Insti