MGMT2700 Harvard How Household Portfolio Diversification Vary With Financial Literacy & Advice? Guidelines (Article Attached): Create a 2-3 page review of

MGMT2700 Harvard How Household Portfolio Diversification Vary With Financial Literacy & Advice? Guidelines (Article Attached):

Create a 2-3 page review of the article.

The review should summarize the article’s key findings as well as the authors’ methodology. Each article review should then briefly describe the strengths and weaknesses of the article authors’ methodology.

Your paper should contain the following sections:

1. Key Findings (what did the authors find?)

2. Methodology Review (what research methods did the authors use to reach their conclusions?)

3. Application (what are some ways in which the authors’ findings are relevant to a finance practitioner?)

Note: Must be plagiarism free. Hans-Martin von Gaudecker
How does Household Portfolio
Diversification Vary with Financial
Literacy and Financial Advice?
DP 03/2011-026 (revised version December 7, 2013)
How does household portfolio diversification vary with
financial literacy and financial advice?
Journal of Finance, forthcoming
Hans-Martin von Gaudecker
?
December 7, 2013
ABSTRACT
Household investment mistakes are an important concern for researchers and policymakers alike; portfolio underdiversification ranks among those mistakes that are
potentially most costly. However, its roots and empirical importance are poorly understood. I estimate quantitatively meaningful diversification statistics and investigate
their relationship with key variables. Nearly all households that either score high on
financial literacy; or rely on professionals or private contacts for advice; achieve reasonable investment outcomes. Compared to these groups, households with below-median
financial literacy that trust their own decision-making capabilities, lose an expected
50bp on average. All group differences stem from the top of the loss distribution.
JEL codes: D14, D12, G11
Keywords: household portfolios, diversification, financial literacy, financial advice.
Universität Bonn, Department of Economics, Adenauerallee 24-42, 53012 Bonn; Germany;
hmgaudecker[at]uni-bonn[dot]de; tel. +49 228 73 9357. Several of the data manipulations and preliminary estimations were performed by Mark Prins for his MSc Thesis at the VU University Amsterdam
and in a subsequent research assistantship to the author. Special thanks to him for an excellent programming
job and many fruitful discussions. Funding from Netspar is gratefully acknowledged. I appreciate helpful
comments received from seminar and conference participants at the Universities of Frankfurt and of Bonn,
at the 2nd SAVE conference in Deidesheim, at the 2011 Netspar Pension Workshop in Amsterdam, and
at the Federal Reserve Bank of Chicago; in particular Tabea Bucher-Koenen, Armin Falk, Dimitris Georgarakos, Michael Haliassos, Rainer Haselmann, Olivia Mitchell, Karl Scholz, Stephen Zeldes, and Michael
Ziegelmeyer. Furthermore, I would like to thank CentERdata for help on numerous data questions; Mauro
Mastrogiacomo for passing on code to prepare the DHS data; and Maarten van Rooij, Annamaria Lusardi,
and Rob Alessie for providing the data and code used to construct the financial literacy scores in Rooij,
Lusardi, and Alessie (2011).
?
1
Economic theory predicts that households will hold their risky assets in the form of a
well-diversified portfolio. The extent to which this prediction holds true has important implications for the regulation of consumer financial products, the design of retirement savings
plans and the distribution of consumer well-being in general. For these issues, it is especially
significant to know the patterns of underdiversification. If only households that are particularly savvy in financial matters invested their retirement wealth in a few highly correlated
stocks, most policymakers would conclude that this was a rational choice. For example, it
could reflect superior information (Nieuwerburgh and Veldkamp, 2010). If it were the least
financially literate doing the same thing, however, it would be much more likely that these
households were simply making poor decisions. In a similar vein, it makes a big difference for
a regulator’s strategy whether the underdiversified households rely on professional financial
advisers; or whether they make their investment choices autonomously.
Only little guidance has come from empirical studies, which can be roughly divided into
two strands. One strand has studied detailed diversification properties using administrative
data from Scandinavian countries (Calvet, Campbell, and Sodini, 2007, 2009; Grinblatt,
Keloharju, and Linnainmaa, 2011). These studies show that—once one accounts for mutual
funds on top of stock portfolios—most households reach reasonable investment outcomes,
but some may expect a very low rate of return given the amount of risk they take. However,
the lack of data on financial literacy and financial advice prevents one from going beyond unconditional distributions of diversification losses with respect to the questions raised above.
A second strand has employed simple measures of portfolio diversification drawn from household surveys (Bilias, Georgarakos, and Haliassos, 2009; Kimball and Shumway, 2010; Guiso
and Jappelli, 2009; Graham, Harvey, and Huang, 2009). In these studies, the data issues
are reversed: While covariates are abundant, the diversification properties do not permit a
quantitative analysis.
In this paper, I combine the strengths of both approaches and calculate quantitatively
meaningful diversification measures for the respondents to a Dutch household survey. Next
to an abundance of background variables, the survey also includes detailed information on
financial literacy and on the most important source of financial advice. In my analysis, I
employ the return loss (Calvet, Campbell, and Sodini, 2007) as the diversification measure.
The return loss is the difference between the maximum expected return that is attainable
at a given standard deviation and the actual expected return for a particular portfolio.
2
The unconditional distribution of the annual return loss reveals that it is limited to half a
percentage point or less up to the fourth quintile. In its top quintile, it averages 1.8 percentage
points, which is large and comparable to previous results.
My main analyses uncover an important interaction of financial literacy and financial
advice in the decision-making process. The return loss of households that seek advice either
in their private network or from professionals does not vary with financial literacy. Among
households making autonomous decisions, the distribution of the return loss looks very similar only if they are endowed with the maximum level of financial literacy. In this group,
a decrease in financial literacy of one standard deviation is associated with an increase in
the return loss by 0.7 percentage points on average. I show by means of quantile regressions
that these effects stem entirely from very large effects at the upper end of the return loss
distribution.
Intrigued by the fact that the outcomes of households relying on professional financial
advisers and on recommendations from within their private network look so similar, I further
investigate the differences between these groups. The use of professional financial advice is
associated with an increase in fees paid across the entire portfolio of 0.3 percentage points.
However, the higher fees do not translate into a significantly higher return loss.
This paper is structured as follows. In the first section, I describe the dataset and
the mean-standard deviation characteristics of households’ portfolios, including the return
loss. In Section II, I explore the sources of underdiversification and scrutinize the effects of
professional advisers, before discussing these results. Section III concludes.
I. The properties of households’ portfolios
Our understanding of the empirical characteristics of households portfolios is based on three
distinct sources of data: (1) tax registers from Scandinavian countries; (2) online brokers’
databases; and (3) household surveys. Each has its own advantages and drawbacks: The
first two sources usually yield extraordinarily detailed portfolio data. However, they do so
for highly selective sets of countries and individuals, respectively. The availability of certain
covariates is another issue. In surveys the situation is the opposite: While representative
samples with a plethora of background information are available all over the world, informa-
3
tion on household portfolios is rather coarse.1
In this study, I combine several strengths of the various approaches by constructing
detailed portfolios for the respondents to the Dutch Central Bank Household Survey (DHS).
Linking individual portfolio components to historical return series allows me to calculate
diversification statistics that are measured in meaningful economic quantities, mirroring
studies based on administrative registers or discount broker data. This section establishes
the usefulness of this approach; Section II goes beyond previous studies by exploiting the
background characteristics present in the DHS data.
A. The DHS data on household portfolios
I use data from the CentERpanel, a Dutch household survey that is administered via the
Internet. In order to avoid selection problems due to lack of Internet access, respondents
without a computer are equipped with a set-top box for their television set (and with a
TV if they do not have one). Respondents are reimbursed for the cost of using the Internet. The CentERpanel is representative of the Dutch population in terms of observable
characteristics. It hosts the Dutch Central Bank Household Survey (DHS), which contains
particularly detailed information on financial matters. For this reason, the data have been
used extensively to describe households’ portfolio choice behavior. Alessie, Hochgürtel, and
Soest (2002, 2004, 2006) and Dimmock and Kouwenberg (2010); and Korniotis and Kumar
(2011) are excellent examples. The CentERpanel was the role model for the RAND American Life Panel, which has emerged as another workhorse in the area of household financial
decision-making (Lusardi and Mitchell, 2007; Hung, Parker, and Yoong, 2009; Hung and
Yoong, 2010).
My analysis is cross-sectional, but in order to increase the sample size I make use of two
waves (2005 and 2006) and adjust all standard errors for clustering within households using
standard methods (StataCorp, 2013). Throughout the analysis, I exclude households with
less than 1,000 euros in financial assets (8.6% of the sample), which yields 1,604 households
to begin with. About a third of these hold risky financial assets, defined as shares, various
types of funds, bonds, and options. See Section A.2 (in particular Table A.1) in the Online
Appendix for more details.
1
A detailed review of the literature and of data sources can be found in the Online Appendix, Section A.1.
4
A unique feature of the dataset is that individuals are not only asked for the number of
stocks and mutual funds they possess, but also to report each item’s name and the quantity
held. I follow the same strategy as Calvet, Campbell, and Sodini (2007, 2009) and connect
each item to its time series of returns obtained from Datastream or Morningstar. This allows
me to estimate the risk-return characteristics of households’ portfolios directly from the data.
Importantly, I do not need to invoke any assumptions about the properties of asset classes.
Furthermore, I can take the covariances between different items’ returns into account in an
appropriate way.
While most households’ reports of their portfolio constituents appear to be very thorough,
I lose about a fourth of the risky financial asset holders because of incomplete reports.
Fortunately, I cannot detect any bias in terms of observable characteristics when dropping
these households. See Tables A.2 and A.3 in the Online Appendix for all details. My final
sample consists of 381 observations.
The households in my sample possess 269 different assets comprised of 99 shares and 170
funds. I observe these assets for more than 11 years on average. The lengths of these return
series vary widely, ranging from below six years at the fifth percentile to almost 20 years
at the ninety-fifth percentile. I conduct all analyses at a monthly frequency, but report
annualized results for easier comparability with the literature.
Of the 170 funds, 106 are equity funds; there are about a dozen funds each investing
in sovereign bonds, corporate bonds, and real estate; 28 funds follow a mixed strategy. I
also obtained data on the fees of these funds. These are about 130 bp on average, with the
fifth and ninety-fifth percentiles at 33 bp and 187 bp. Section A.3 in the Online Appendix
describes the characteristics of the portfolios’ components in more detail.
B. The mean-variance characteristics of households’ portfolios
I follow Calvet, Campbell, and Sodini (2007) and assume that assets are priced according to
an international CAPM. Directly estimating the expected return of each household’s portfolio
would be problematic because of the short return histories for some assets; and because the
time series cover different time spans.2 I take the MSCI Europe Index to proxy the efficient
2
To see this, assume that one observes two assets with identical moments. Data for the first is available
in the 2000-2005 period and for the second in the 2002-2007 period. The first asset would likely have a
much lower estimated ? because the market conditions were worse during the earlier period. Pricing assets
5
market portfolio, which is a natural choice for a member of the Eurozone; results are robust
to using the AEX or the MSCI World Index instead. All returns are framed as excess returns
over the risk-free rate, which is approximated by the one-month EURIBOR. Before applying
the CAPM, I subtract mutual funds’ fees from their gross returns and 30 basis points from
the benchmark index’s returns. This is approximately the cost charged by index funds
replicating common benchmarks.
Net of this “fee”, the MSCI Europe has an annual excess return of µb = 5.77% in the
period of January 1983 to July 2009. Along with its standard deviation of ?b = 16.7%,
this implies a Sharpe ratio Sb =
µb/?
b
of 35%. Imposing the CAPM leads to the following
regression for all assets a ? {1, 2, . . . , 269}:
e
e
ra,t
= ?a · rb,t
+ ?a,t .
(1)
Given the ?s of all assets and the portfolio weights for each household, it is straightforward
to calculate the expected return µh of each household’s portfolio.
In order to gain an understanding of the basic characteristics of household portfolios, it is
useful to plot them in the mean-standard deviation plane. Panel A of Figure 1 does this for
the pure stock portfolios and reveals a picture of strong underdiversification, which is similar
to the findings of Calvet, Campbell, and Sodini (2007) for Sweden and Goetzmann and
Kumar (2008) for the United States. The mutual fund component of households’ portfolios
appears to be much better diversified. The major part of the distribution lines up right
below the efficient frontier (Panel B of Figure 1). Nevertheless, a substantial fraction of
mutual fund portfolios perform significantly worse than the market portfolio at any risk
level. Panel C contains the aggregate of stocks and mutual funds and shows that many
households reduce the risk of their stock portfolios by additionally investing in mutual funds
(of all risky asset owners, 55% only own mutual funds, 18% only own stocks, and 26% own
both). Panel C implies that studies forced to focus on stock portfolios (e.g., Grinblatt,
Keloharju, and Linnainmaa, 2012; Goetzmann and Kumar, 2008) miss an important part of
the picture.
The distance to the efficient frontier shrinks a lot more when holdings of safe assets are
via the CAPM avoids this problem as long as the assets’ correlation structure does not change with market
conditions.
6
Figure 1. The mean-standard deviation characteristics of household portfolios
2.5
5
Return mean (%)
7.5
10 12.5
15
17.5
B: Mutual funds portfolios
0
0
2.5
5
Return mean (%)
7.5
10 12.5
15
17.5
A: Stocks portfolios
0
10
20
30
Return standard deviation (%)
40
50
0
20
30
Return standard deviation (%)
40
50
40
50
2.5
5
Return mean (%)
7.5
10 12.5
15
17.5
D: Financial portfolios
0
0
2.5
5
Return mean (%)
7.5
10 12.5
15
17.5
C: Risky portfolios
10
0
10
20
30
Return standard deviation (%)
40
50
0
10
20
30
Return standard deviation (%)
Source: CentERpanel, Datastream, Euroinvestor, own calculations.
taken into account in Panel D. Nevertheless, a few outliers with severe losses remain at high
levels of risk; a number of households hold portfolios one to two percentage points below the
efficient frontier at relatively low levels of risk. Diversification losses of this magnitude will
be substantial when accumulated over the life-cycle (Calvet, Campbell, and Sodini, 2007;
Tang, Mitchell, Mottola, and Utkus, 2010). Detailed descriptions of mutual fund fees and
of other characteristics of the funds and shares in my sample are relegated to Section A.3 in
the Online Appendix.
7
C. The expected costs of underdiversification
A variety of measures are available to reduce the two-dimensional graphs in Figure 1 to a
single statistic of diversification losses.3 The most comprehensive criterion is the return loss,
defined by Calvet, Campbell, and Sodini (2007) as the expected return a household loses by
not choosing a position on the efficient frontier with the same level of risk as its portfolio.
In graphical terms, the return loss is the vertical distance between the efficient frontier and
the location of a household’s portfolio in Panel D of Figure 1.
The first column in Table I shows the average return loss within each quintile of its
distribution. Return losses are limited to half a percentage point per year up to the fourth
quintile; they reach 1.8 percentage points annually in the highest quintile. As mentioned
before, such a number implies a large reduction in wealth when accumulated over the life
cycle.
In order to get an idea of what contributes to the return loss, it is useful to write it as the
product of the expected excess return on the market portfolio (which does not vary across
households), the risky asset share ?h , ?h , and the relative difference between the Sharpe
ratios of the efficient market portfolio and the household’s portfolio:
RLh = µb · ?h · ?h ·

Sb ? Sh
Sh

.
(2)
High values of ?h and ?h may be signs of efficient risk-taking; the quantity (Sb ? Sh )/Sh unambiguously measures a loss from underdiversification.
The remaining columns of Table I contain the three terms that vary across households.
Of course, the different components do not add up due to Jensen’s inequality; hence this
illustration does not qualify as a decomposition. Nevertheless, it remains useful to get a
rough idea of the underlying magnitudes. The risky asset share increases quickly until the
middle of the return loss distribution and increases modestly afterwards. The ? coefficient
rises almost linearly over the quintiles, so inefficient and efficient risk-taking at least go hand
in hand on average. The diversification loss is fairly constant in the lower quintiles and
increases strongly in the top quintile. It is the only one of the three components of RLh
that is significantly different between the fourth and fifth quintiles of RLh at the 10%-level
3
Section B in the Online Appendix describes a number of these measures and provides a detailed comparison to the analysis of Calvet, Campbell, and Sodini (2007).
8
Table I
Return loss and its components by quintile
Return loss Return Fraction in
quintile
loss
risky assets
RLh
?h
Lowest
Second
Third
Fourth
Highest
0.04
(0.00)
0.14
(0.00)
0.28
(0.01)
0.52
(0.01)
1.80
(0.16)
0.08
(0.01)
0.23
(0.02)
0.43
(0.03)
0.54
(0.03)
0.60
(0.03)
Beta
coefficient
?h
Diversification
loss?
0.77
(0.04)
0.83
(0.06)
0.89
(0.04)
0.91
(0.04)
1.01
(0.07)
0.30
(0.08)
0.46
(0.10)
0.27
(0.04)
0.50
(0.10)
0.88
(0.12)
Sb ?Sh
Sh
Source: CentERpanel, Datastream, Euroinvestor, own calculations. The table entries are average values
of the statistics denoted in the table header by return loss quintile. Standard errors of the means are in
parentheses.
?
The value (Sb ? Sh )/Sh becomes extremely high if the expected return on a household’s portfolio µh is close
to zero. I therefore winsorize (Sb ? Sh )/Sh above at its ninety-fifth percentile.
(p-…
Purchase answer to see full
attachment