- 1 -
International diversification through country index funds and ETFs
By Yan Zhao♦
First draft Sept, 2007
This draft: Aug, 2008
Abstract
This paper studies the international asset allocation by comparing various portfolio strategies
on country index funds and ETFs. We compare the performance of simple portfolio
construction strategies, such as the equally-weighted, capitalization-weighted, GDP-weighted
and dividend-weighted portfolios, with optimal portfolio strategies based on mean-variance
analysis: the global minimum variance portfolio, the mean-variance efficient portfolio and the
Bayes-Stein shrinkage portfolio.
We find simple portfolio strategies perform similarly to optimal portfolio strategies based on
mean-variance analysis in developed countries while outperform optimal portfolio strategies
in emerging markets. Which shows the loss from simple rather than optimal
diversification is smaller than the loss from model estimation errors. Among the simple
portfolio strategies, the most commonly used capitalization-weighted portfolio suffers from
price fluctuations and systematically over-weights the over-valued countries and
under-weights the under-valued countries; to overcome this shortcoming, we propose the
portfolio based on the GDP weights since the GDP is the most straightforward fundamental at
the country level and a relative stable measure of the size of each country in the global
economy. A delta strategy based on the difference of a country’s GDP share and capitalization
share has predictive power. We further show the gains from diversification can be achieved in
portfolios consisting of as few as 3 developed countries or 4 emerging markets.
♦ International Business School, Brandeis University, [email protected]
- 2 -
1. Introduction
Mean-variance efficient portfolios play an important role in portfolio allocation
problems. An investor who cares only about the mean and variance of portfolio
returns should hold an optimal portfolio on the mean-variance efficient frontier
(Markowitz (1952)). To implement this optimal portfolio in practice, an investor has
to estimate the mean and variance of a portfolio’s returns via their sample analogues.
However, due to estimation errors portfolios constructed using these estimators are
extremely unstable and the optimal weights on each asset fluctuate substantially over
time, and thus these unstable portfolios perform poorly out of sample (Littlemand
(2003), Brandt (2004), Roon(2004), DeMiguel, Garlappi and Uppal (2007), and
Demiguel and Nogales (2007)).
The instability of the mean-variance portfolios can be explained partly by the
well-documented difficulties associated with estimating mean asset returns. For
instance, Jagannatham and Ma (2003) state that ‘the estimation error in the sample
mean is so large that nothing much is lost in ignoring the mean altogether’. For this
reason, the global minimum variance portfolio, which relies solely on estimates of the
covariance matrix and is not as sensitive to estimation errors, also plays an important
role in asset allocation decision.
Another prominent role in the asset allocation decision is played by the shrinkage
estimation approach. The idea of shrinkage estimation is attributed to James and Stein
(1961) and is designed to handle errors in estimating expected returns by shrinking the
sample means toward a grand mean. The estimator thereby reduces the extreme
estimation errors that may occur in the cross section of individual means.
Despite considerable efforts to handle the estimation errors, the out-of-sample
performance of portfolios based on mean-variance analysis is still not consistently
better than a simple equally-weighted portfolio, for instance, DeMiguel, Garlappi and
Uppal (2007) compare fourteen models of optimal portfolio choice using seven
empirical datasets, and find none of them consistently outperform the equally-
weighted portfolio. Thus, keeping portfolio strategies simple is very important
empirically.
- 3 -
The most widely used simple portfolio strategy is the capitalization-weighted
portfolios, such as the MSCI world index and MSCI EAFA index, which is well
accepted by practitioners and academics. The benefits of the traditional capitalization
weighted index are numerous (Schoenfeld (2006), Arnott, Hsu, and Moore (2005)): it
is self-rebalanced, and thus incurs low trading costs, it has high investment capacity
and by design it is a market clear investment. However, the capitalization-weighted
portfolio is flawed in a fundamental way: it systematically over-weights the
over-valued stocks and under-weights the under-valued stocks as stock markets
fluctuate;
A neutral way to overcome these shortcomings is instead to use GDP as the weighting
scheme. Not only the GDP is the most straightforward fundamental at the country
level, but also it reflects the size of each country in the global economy. A country’s
GDP is highly correlated with the market size and is a relatively more stable measure.
As documented in Hamza, Kortas, LHer and Roberge (2005, 2006), the GDP is a
better weighting scheme than capitalization when investing globally.
Other weighting schemes, such as the dividend weighting scheme is inspired by
Arnott, Hsu and Moore (2005), who construct portfolios based on metrics of firms’
book values, revenues, dividends, incomes and others. Using data for the U.S data
from 1962 to 2004, they report the portfolio weighted by dividends outperform the
S&P500 in a variety of scenarios. Hsu and Campollo (2005) show that portfolios
based on dividends also outperform the capitalization-weighted portfolio in 23
developed countries besides the U.S from 1984 to 2004. Estrada (2006) further shows
the superior performance of the dividend-weighted portfolio in 16 developed
countries over a period from 1973-2006.
The objective of this paper is to investigate the portfolio construction issue at country
level. We compare the performance of the above optimal portfolio strategies based on
mean-variance analysis: the mean-variance efficient portfolio, the global minimum
variance portfolio, and the Bayes-Stein shrinkage portfolio with the simple portfolio
strategies of investing in each of the countries equally or according to their relative
shares in the capitalization, GDP, or dividends using 43 country index funds and
ETFs.
- 4 -
By comparing the above different portfolio strategies, we have the following findings.
First, we find simple portfolio strategies perform similarly to, if not better than, the
optimal portfolio strategies based on mean-variance analysis in developed countries
while outperform the optimal portfolio strategies in emerging markets. Which shows
the gain from optimal diversification is offset by estimation errors, and the loss from
simple rather than optimal diversification is smaller than the loss from model
estimation errors.
Secondly, among the simple portfolio strategies, the most commonly used
capitalization-weighted portfolio suffers from price fluctuations and systematically
over-weights the over-valued countries and under-weights the under-valued countries.
To overcome this shortcoming, we propose a portfolio weighting scheme based on the
GDP weights. Our empirical results illustrate that GDP-weighted portfolios
outperform capitalization-weighted portfolios. We claim that it is because GDP is the
most straightforward fundamental at the country level and measures the size of each
country in the global economy.
Thirdly, we further show the gains from diversification can be achieved in portfolios
consisting of as few as 3 developed countries or 4 emerging markets. In the case of
portfolios consisting of a smaller number of countries, the estimation errors are less
severe and thus, the optimal portfolios have better out of sample performances.
Finally, we propose on the one hand, GDP reflects a country’s fundamentals and is
more stable, on the other hand, a country’s market capitalization fluctuates due to
various noises and it should come back to fundamentals sooner or later. We then claim
that a delta strategy which puts delta weight (delta weight=GDP weight – CAP weight)
on each country will have some predictive power. If the delta weight is positive, we
believe the country’s ETF is undervalued and will buy delta shares of that country,
similarly, if delta weight is negative, which shows the country’s ETF is overvalued,
and we will short delta shares of that country. This long-short market neutral strategy
works well in mean-reverting markets like emerging markets, while does not work in
a trending market like developed countries in 1980.
- 5 -
The remainder of the paper is organized as follows. Section 2 describes the data and
methodology. Section 3 reports the empirical performance of different portfolios.
Section 4 is the robustness check and Section 5 concludes.
2. Sample selection, portfolio strategies and evaluating methodology
2.1 Data
Quarterly nominal GDP in dollar terms and monthly dividend and market
capitalization data are taken from Datastream. Monthly return series are calculated
using Morgan Stanley Capital International (MSCI) indexes in dollar terms, and
accounting for both capital gains and dividends. Finally, the U.S. 3-month T-bill rates
(risk-free rates) are also from the Datastream. The sample period runs from Dec 1979
to Dec 2006 for 23 developed countries, and from Dec 1994 to Dec 2006 for 20
developing countries. The sample periods are selected to cover as long a history as
possible with data from Datastream database.
Table 1 and table 2 show the summary statistics of the return series for both 23
developed countries and 20 emerging markets. The distributions of returns series are
quite different in developed countries and in emerging markets. In general, emerging
markets have higher volatilities and higher excess kurtosis, suggesting that the return
distributions are more fat-tailed than a normal distribution and more outliers.
2.2 Mean-variance portfolio strategies
2.2.1 Mean-variance efficient portfolio (MVE)
In the mean-variance framework of Markowitz (1952), we consider an investor whose
preferences are fully captured by the mean and variance of a portfolio. At each time t,
the investor selects tω to maximize the following utility subject to no short-selling
constraints; the no short-selling constraints are mainly for the purpose of reducing the
extreme weights on each country:
Max ttttt ωωγ
µω Σ− ''
2 subject to 11' =tNω and 0>=itω ………………………..(*)
Where tµ denotes the N-vector of expected excess returns (over the risk free asset)
on the N risky assets available for investors at time t. tΣ is the corresponding N x N
variance matrix of returns. tω is the vector of portfolio weights invested in the N
- 6 -
risky assets at time t. N1 defines a N x1 vector of ones. Finally, γ denotes an
investor’s risk aversion, in our following analysis, γ is chosen to be 2, results for
other values of γ are discussed in robustness check.
To implement this classic model, we solve the above maximization problem with the
mean and covariance matrix of asset returns replaced by their sample counterparts,
t
^
µ and t
^
Σ , respectively. The sample mean and covariance matrix is calculated using
all the observations up to time t, that is, we do not drop the earliest observation when
we add a new observation. Note that this portfolio strategy completely ignores the
estimation errors.
2.2.2 Bayes-Stein shrinkage portfolio (BS)
Shrinkage estimation is pioneered by Stein (1955) and James and Stein (1961), the
shrinkage estimator shrinks the sample means toward a grand mean. In our analysis,
we follow the estimator proposed by Jorion (1985, 1986), who takes the grand mean
to be the mean of the global minimum variance portfolio. In addition to shrinking the
estimate of the mean, Jorion also accounts for estimation error in the covariance
matrix through traditional Bayesian estimation method. More specifically, following
Jorion (1986), the estimator we use for expected returns is
MIN
tttt
BS
t
^^^^^
)1( µφµφµ +−=
Where
)()'()2(
2
^^1^^^
^
MIN
ttt
MIN
tt
t
MN
N
µµµµ
φ
−Σ−++
+=
−
For 10^
<< tφ the Jorion estimator shrinks the sample means toward the mean of the
global minimum variance portfolio MIN
t
^
µ . The estimator thereby reduces the extreme
estimation errors that may occur in the cross-section of individual means. The
implement in Jorion (1986) also accounts for estimation errors in the covariance
matrix by a traditional Bayesian estimation. The following expression for the
covariance matrix is utilized in the portfolio construction.
NtN
NN
t
t
t
t
BS
t
MMM 11
11
)1(
)1
1(1^
'
'
^
^
^
^^
−
Σ++
+
+
+Σ=Σ
τ
τ
τ
- 7 -
t
tt M
^
^
^
1 φ
φτ
−
= Where M is the length of estimation window up to time t.
The Bayes-Stein portfolio is constructed by maximize the utility function (*) in
section 2.2.1 using BS
t
^
µ and
BS
t
^
Σ , which combines a shrinkage approach and a
traditional Bayesian estimation, and hence, is known as the Bayes-Stein portfolio.
2.2.3 Global minimum variance portfolio (MIN)
The global minimum variance portfolio of risky assets is constructed to minimize the
variance of the portfolio variance subject to no short-sale constraints, that is,
Min ttt ωω Σ' subject to 11' =tNω and 0>=itω
To implement this strategy we only use the estimate of the covariance matrix of asset
returns, and in this paper, we use the sample covariance matrix using the observation
up to time t and completely ignores estimates of the expected returns.
2.3 Simple portfolio strategies
The most straight forward simple strategy is to hold a portfolio that equally invests in
each of the countries. We call it equally-weighted portfolio (EQUAL). This strategy
completely ignores all estimates of the data. A second simple strategy is the
capitalization-weighted portfolio (CAP). To construct the capitalization-weighted
portfolio, the weight on each country is set according to their relative market
capitalization at time t, the relative market capitalization is calculated as the market
capitalization of each country relative to the sum of the capitalization of all countries.
In a similar fashion we also construct the GDP-weighted portfolio (GDP) and the
dividend-weighted portfolio (DIV). We finally examine a composite strategy by
equally weighting the CAP, GDP and DIV weights (COMBO). To calculate the GDP,
CAP and DIV weights, we use the trailing 5-year averages to reduce the excessive
volatility in portfolio weights would result from using current data. For the GDP data,
we use the data lagged by a year to account for all the seasonally adjustment and
make sure the data are available at the time of portfolio rebalance. The
capitalization-weighted portfolio bears close resemblance to the MSCI benchmarks,
but they are not identical, the MSCI benchmark indices are weighted by float not
- 8 -
aggregate capitalization and are rebalanced annually. In addition, the
capitalization-weighted portfolios in this paper use trailing 5-year average
capitalization to reduce the excessive volatility and are rebalanced quarterly.
Finally, all the portfolios are rebalanced at the end of each quarter, and returns for
these portfolios are calculated on a monthly basis. For reference, the performance of
the MSCI world index and MSCI emerging markets index in the same sample period
are also reported.
2.4 Methodology for evaluating performance
The measurements for evaluating performances of different portfolios are the
followings:
Alpha is calculated as ))(()( fmft RREbetaRRE −−− using the CAPM model,
where MSCI index is used as a proxy of the market return, and risk-free rate is the
annualized U.S. 3-month T-bill rate.
Sharpe ratio is defined as excess returns over the risk-free rates divided by their
standard deviation. In formula: )( t
ft
RStd
RRoSharpeRati
−=
Certainty-equivalent return (CEQ) is the risk-free rate that an investor is willing to
accept rather than adopting a risky portfolio strategy, we compute the CEQ as the
following: VarMeanCEQ2
γ−= , where Mean and Var are the mean and variance of
the portfolio return over risk-free rates. γ is the risk aversion of investors, which is
set equal to 2 in our analysis.
Turnover rate is the amount of trading required to implement each portfolio strategy.
We compute the portfolio turnover, defined as the average sum of the absolute value
of the trades across all available assets, as indicated by the following formula:
|))1()((|1
1
'
1
+−= ∑∑==
twtwT
Turnover i
T
t
i
T
t
Where )1( +twi is the weight at time t+1; while )(' twi is the weight just before the
rebalance, it is different from the weight at time t, because changes in asset prices
have caused a change in the relative weights in the portfolio
- 9 -
In addition to reporting the raw turnover rates for each strategy we also report the
transaction cost generated by the turnover rate. Following Balduzzi and Lynch
(1999), DeMiguel, Garlappi and Uppal (2007) , we assume a 50 basis points
transaction cost per one-way transaction, the total transaction cost is calculated as
turnover*0.5%*2.
3. Empirical Results
3.1 Performance comparison
The performance comparison is done separately in developed countries and emerging
markets. Two simple reasons that we do not construct portfolios in a combined
universe; the first one is that the sample periods for developed countries (1979-2006)
and emerging markets (1994-2006) are different, significant data would be lost if we
combine the two universes. Secondly, the weights on emerging markets will be
overwhelmed by the weights on developed countries in the capitalization-weighted
and GDP-weighted portfolios.
3.1.1 Developed countries
In this section, we compare empirically the performance of portfolios constructed
based on the mean-variance analysis with portfolios based on simple strategies listed
in the above section. Table 3 reports the performance of all portfolios for 23
developed countries, and Figure 1 shows the performance of these portfolios in a
mean-variance space.
To compare the performance of different strategies, we first look at the Sharpe ratio of
different portfolios. By design, the ex-post mean-variance efficient portfolio should
have the highest Sharpe ratio when there are no estimation errors. The magnitude of
the difference between the ex-post mean-variance efficient Sharpe ratio and the
ex-ante mean-variance efficient Sharpe ratio measures the loss from model estimation
errors when implementing the mean-variance strategy. Similarly, the difference
between the ex-post mean-variance efficient Sharpe ratio and the Sharpe ratios of
simple portfolio strategies measures the loss from simple rather than optimal
diversification when there are no estimation errors.
We first look at the loss from model estimation errors. We can see the difference
- 10 -
between ex-post mean-variance efficient Sharpe ratio and ex-ante efficient Sharpe
ratio, as indicated in Table 3, is as big as 0.57 (1.07-0.50), which means an investor
cannot fully capture the theoretical gains from optimizing models due to estimation
errors. The poor ex-ante performance of mean-variance strategies is because the
ex-ante portfolio weights vary substantially from the ex-post optimal weights, as
shown in table 6. While the ex-ante global minimum variance portfolio seems to have
a better Sharpe ratio compared to the ex-ante mean-variance efficient Sharpe ratio
(0.73 vs. 0.50), which shows that by ignoring the estimates of expected returns
altogether but exploiting the information about correlations only does lead to better
performance. Since the ex-ante mean-variance efficient portfolio completely ignores
the estimation errors, the Bayes-Stein portfolio which explicitly accounts for
estimation errors might also lead to a better performance in terms of Sharpe ratio. And
indeed, the Sharpe ratio of the Bayes-Stein portfolio is 0.72, which is also higher than
the ex-ante mean-variance efficient Sharpe ratio of 0.50.
Other performance measures, like Portfolio Alpha and CEQ, tell a similar story as
above. By construction, the ex-post mean-variance efficient portfolio has the highest
Alpha and CEQ, but ex-ante, they all suffer greatly from estimation errors. For
instance, the CEQ of the mean-variance efficient portfolio drops from 21.41% ex-post
to 12.13% ex-ante, which is even lower than the CEQs of all simple strategies (except
for the capitalization-weighted portfolio).
Secondly, to look at the loss from simple rather than optimal diversification when
there are no estimation errors we compare the difference between ex-post
mean-variance efficient Sharpe ratio and the Sharpe ratios form simple portfolio
strategies. For example, the Sharpe ratio of the equally weighted portfolio is 0.75,
indicating a 0.32 loss (1.07-0.75).
And in general, simple strategy portfolios have Sharpe ratios in a range of 0.49-0.75,
while the ex-ante optimal portfolios have Sharpe ratios in a range of 0.50-0.73. This
similar range of the Shape ratio shows that simple portfolio strategies are comparable
to if not better than optimal strategies based on the mean-variance analysis because
they do not suffer from the model estimation errors.
- 11 -
Finally, if we think of practical issues when implementing these different portfolios,
an importance concern is the portfolio turnover rates, which directly affect transaction
costs and net returns. From Table 3 we can see that the turnover rates of all ex-ante
optimal portfolios are much higher than the simple portfolio strategies, since those
ex-ante optimal portfolio weights are extremely sensitive to model inputs.
The various performance measures analyzed above illustrate that simple portfolios
suffer less from model estimation errors and have lower turnover rates, thus to keep
portfolio strategy simple is very important empirically. If we take a close look at the
different simple strategies, we can notice that the most commonly used capitalization-
weighted portfolio delivers the lowest Alpha, Sharpe ratio, CEQ and net returns
among all strategies. Since all five simple strategies consist of the same 23 MSCI
country ETFs, the difference in the performance stems solely from the different
weights given to these benchmarks in each ETF. Table 6 presents the average weights
on each country.
The first explanation of the better performance of other simple portfolios over the
capitalization-weighted portfolio is that the capitalization-weighted portfolio can be
highly concentrated when one major country or region outperforms others for a
sustained period of time. For instance, during the 1980s Japan market bubbles, the
capitalization-weighted portfolio put around 40% weights on Japan from 1989 to 1994,
as showed in the figure 3. And when the Japan bubbles burst in late 1990s, the
performance of the capitalization-weighted portfolios drops. To exclude the Japan
effect, we reconstruct all portfolios using 22 countries excluding Japan, the results are
show in Table 4. Table 4 shows once we exclude Japan, the performance of the
capitalization-weighted portfolio improves, for instance, alpha increases to 2.03%
excluding Japan from -0.17% including Japan; and Sharpe ratio increases to 0.62 from
0.49. A second explanation of the better performance of other simple portfolios over
the capitalization-weighted portfolio is that they are less affected by the price
fluctuations, since the capitalization-weighted portfolio systemically over-weights the
over-valued assets and under-weights the under-valued assets as prices fluctuate;
that’s why once we controlled the Japan effect, the capitalization-weighted portfolio
still underperforms other simple portfolios.
- 12 -
A neutral way to overcome the shortcomings of the capitalization-weighted portfolio
is instead to use GDP as the weighting scheme. Not only the GDP is the most
straightforward fundamental at the country level, but also it reflects the size of each
country in the global economy. In addition, GDP is a relatively more stable measure.
As show in figure 3, while the capitalization-weighted portfolio highly concentrated
on Japan in 1980s, the GDP-weighted portfolio gives a very stable weight on Japan
around 15-20%. And indeed, from the different measure of performance, the
GDP-weighted portfolio delivers a higher Alpha, Sharpe ratio and CEQ while
maintaining a similar turnover rate compared to the capitalization-weighted portfolio.
Other weighting schemes also help to overcome the shortcomings of the
capitalization-weighted portfolio, such as portfolio based on dividends paid. However
this strategy must be implemented with cautions, because in some countries,
dividends may not be regularly paid, and investors may not relate dividends to stock
prices closely.
Finally, the equally-weighted portfolio seems to deliver the best alpha, Sharpe and
CEQ at the cost of slightly higher turnover rates among the simple strategies, but
again, cautions must be used when using this strategy because of the small country
effect. As show in table 6, the equally-weighted portfolio systematically puts more
weights on smaller countries where the returns are higher. This superior performance
of the equally-weighted portfolio is consistent to the findings of DeMiguel, Garlappi
and Uppal (2006) and Hamza, Kortas, LHer and Roberge (2005, 2006).
3.1.2 Emerging markets
Stories are getting more interesting when we turn to the emerging markets. Table 5
reports the performances of all portfolios for 20 emerging markets, and Figure 2
shows the performances of these portfolios in a mean-variance space.
We still first take a look at the loss from model estimation errors by comparing Sharpe
ratios of different portfolios; and we can see in general, the loss from estimation erros
is big. The ex-post mean-variance efficient Sharpe ratio is 1.20, while the ex-ante
mean-variance efficient Sharpe ratio is only 0.49, suggesting there are substantial loss
of 0.71 from estimation errors. Again, this poor ex-ante performance is because of the
- 13 -
deviation from the ex-post optimal weights assigned to each county, which is shown
in table 7. The Bayes-Stein portfolio which accounts for estimation error by shrinking
the sample mean to a common grand mean does have improvement with a Sharpe of
0.72. By ignoring the estimates of expected returns altogether but exploiting the
information about correlations, the ex-ante global minimum variance portfolio also
seems to perform very well which has a Sharpe ratio of 0.76.
We now turn to the loss from simple rather than optimal diversification when there are
no estimation errors. Comparing Sharpe ratios of simple strategies to the portfolios
based on mean-variance analysis, we find the simple strategies yield higher Sharpe
ratios compared to ex-ante optimal portfolios; the simple portfolios have Sharpe ratios
in a range of 0.78-0.89, while the ex-ante mean-variance efficient portfolios have
Sharpe ratios in a range of 0.49-0.76. The higher range of Sharpe ratios of simple
strategies shows that errors in estimating means and covariance destroy all the gains
from optimal relative to simple diversification, investors are better off using the
simple strategies.
Other performance measures, like Portfolio Alpha and CEQ, tell a similar story as
above. For instance, the Alpha of the ex-ante mean-variance efficient portfolio is less
than half of the ex-post, and CEQ of the ex-ante mean-variance efficient portfolio is
around one-third of the ex-post. And simple portfolios in general have higher alpha
and CEQ than the ex-ante optimal portfolios. Therefore, the simple portfolios
generally achieve better performances than the ex-ante optimal portfolios based on
mean-variance analysis due to the much smaller loss suffered from model estimation
errors.
Finally, we consider the practical issues of the portfolio turnover rates. The range of
the turnover rates of simple portfolios (71.24% to 85.37%) is actually lower than the
range of ex-ante optimal strategies (84.44% to 95.77%), lower turnover rates thus lead
to higher net returns; which further shows the investors are better off using the simple
strategies. Therefore, to keep portfolio strategy simple is more attractive in emerging
markets.
Compared to other simple portfolios, the superior performance of the GDP-weighted
- 14 -
portfolios comes from different weights put on each country. In emerging markets,
country sizes are positively related to equity market returns, GDP weights
systematically put more weights on bigger countries where returns are higher. It is
shown in table 7, the GDP does assign much greater weights on the high returns
countries, examples are Russia, Brazil and Mexico. In contrast, country sizes are
negatively related to equity market returns among developed countries, the
equally-weighted portfolio works better than other simple portfolios because equally-
weighted portfolio puts relative more weights on small countries where the returns are
high.
Finally, the Dividend-weighted portfolio also achieves good performance. However,
in the emerging markets, due to the fast growing and different culture, companies may
choose to reinvest most of the profits instead of paying dividends to fully capture the
growth opportunity; in addition, because of the culture differences, dividends may be
paid ac hoc. Thus, stock prices in the emerging markets may not closely relate to
dividends; and the fluctuation of dividends may not reflect the company’s
fundamental and hence the dividends might not serve as weighting scheme.
3.2 How many countries are needed to diversify efficiently?
The above performances are all based under assumption of investing in 23 developed
countries and 20 emerging markets. The question is what is the minimum number of
countries needed to be included in a portfolio in order to achieve similar performance?
In this section, we show 3 largest developed countries according to their market
capitalization are efficient to achieve similar performance in Table 3 as discussed
above; and for emerging markets, as few as 7 largest countries in terms of
capitalization or 4 largest countries in terms of GDP are efficient.
To determine how many countries are needed at least to diversify efficiently in
developed countries, we plot portfolio variances against number of countries included
in each portfolio in figure 4. First I rank the countries based on their capitalization
values, the x-axis can be read as: top 1 country in terms of cap (which is the US), top
2 countries in terms of cap (which are the US and Japan), top 3 countries in terms of
cap (the US, Japan and the UK). The y-axis is the equally-weighted,
capitalization-weighted and GDP-weighted portfolio risks (STD). Figure shows the
- 15 -
portfolios consisting 3 countries achieve the lowest portfolio risk.
To further evaluate whether the performance of portfolios consisting of 3 developed
countries is comparable to the performance of portfolios consisting of 23 developed
countries, we report the performance measures in table 8.
The first thing to note is that because of the small number of countries (N=3), the
number of moments to be estimated is small, thus, estimation errors are less severe.
Comparing the ex-ante mean-variance efficient Sharpe ratio and ex-post
mean-variance efficient Sharpe ratio, the loss is 0.12 (0.62-0.49) which is much
smaller than in the case of 23 countries. The global minimum variance portfolio and
the Bayes-Stein shrinkage portfolio achieve Sharpe ratios of 0.52 and 0.56
respectively and the loss from estimation errors is even smaller in those two portfolios.
The CEQ measure shows the same story, the average CEQ of ex-ante optimal
portfolios are about only 1% less than those of the ex-post optimal portfolios
Secondly, in general, the performance of the mean-variance portfolios is slightly
better than those of simple portfolios. This is as expected, as the number of assets
decrease, estimation errors decrease, thus investors are more likely to capture the
gains from optimization.
Finally, although the benefit from diversification is smaller due to the small number of
countries, net returns of different portfolios are still comparable to the net returns in
the case of 23 countries. And therefore, in develop countries the risk-sharing gain can
be achieved in portfolios consisting of as few as three.
We then look at the emerging markets; stories are quite different in emerging markets.
Figure 5 ranks the emerging markets based on their capitalization values, it shows the
largest 7 countries in terms of capitalization achieves the lowest portfolio risk. These
countries are Taiwan, Korea, Brazil, South Africa, India, Mexico and Malaysia.
Similarly, Figure 6 ranks the emerging markets based on their GDP values, it shows
the largest 4 countries in terms of GDP achieves the lowest portfolio risk. These
countries are Brazil, Mexico, Korea and India. Unlike in the case of developed
countries, where the US, the UK and Japan are the largest countries in terms of
- 16 -
capitalization across all time period; in the universe of emerging markets, the largest
countries in terms of capitalization and GDP slightly vary across different time period.
Thus, to make our portfolios implementable, the portfolios include the largest
countries at the time of rebalance.
Table 9 and Table 10 show the performance of the different portfolios in case of 7 and
4 countries. Because of the small number of moments to estimate, the estimation error
problem is less severe and the loss from optimization is less, which is indicated by the
difference of Sharpe ratios between ex-ante and ex-post optimal portfolios. For
instance, the difference of Sharpe ratio between the ex-ante mean-variance efficient
Sharpe ratio and the ex-post mean-variance efficient Sharpe ratio is only 0.17, which
is much smaller than in the case of 20 countries.
Secondly, consistent to the previous results, GDP-weighted portfolios achieve better
performances than the ex-ante optimal portfolios; GDP-weighted portfolios have
higher Sharpe ratios, CEQ, Alphas than the ex-ante optimal portfolios, while at the
same time, maintain lower turnover rates and thus, higher net returns.
And finally, portfolios consisting of as few as 4 or 7 countries achieve comparable
gains from risk-sharing in the case of 20 emerging markets. And actually portfolios
consisting of 4 biggest countries in terms of GDP achieve higher gross returns, alpha,
CEQ, and lower transaction costs, thus higher net returns than the portfolios of 20
emerging markets.
3.3 Delta-strategy
If we believe GDP reflects fundamentals and is stable, and the market capitalization
fluctuates due to noisy prices, but should come back to fundamentals sooner or later.
Then a delta strategy will have some predictive power. The delta strategy puts a delta
weight (=GDP weight – CAP weight) on each country. If the delta weight is positive,
we believe the country’s ETFs is undervalued and will buy delta shares of that country,
similarly, if delta weight is negative, which shows the country’s ETFs is overvalued,
and we will short delta shares of that country. Thus, the delta strategy is a long-short
market neutral strategy.
- 17 -
By construction this delta strategy should work well in emerging markets, where the
GDP weight of a country can be very different from the capitalization weight,
investors can benefit from buying undervalued ETFs at quarter t and selling them at
t+1 when prices of the ETFs grow back to fundamentals and selling overvalued ETFs
at quarter t and buying them back at t+1 when prices of the ETFs fall back to
fundamentals. And indeed, table 11 shows the delta strategy yields a net return of as
high as 29.58% with Sharpe ratio of 1.29 in emerging markets.
However, this delta strategy may not work well in a trending market like the
developed countries in the 1980s, where the prices come back to fundamentals at a
relatively slow speed. As we can see the delta strategy only yield 0.71% gross return
in developed countries, and the Sharpe ratio is negative. This poor performance of the
delta strategy is because the long-side and short-side portfolios grow almost at the
same speed due to the bubble in Japan in 1980s. As shown in figure 6, through out the
period from January 1981 to June 1995, the weights on Japan in the short-side of the
portfolio are always above 40%; and more than 80% from October 1985 to December
1990; in these periods the short-side portfolio grows even faster than the long-side
portfolio due to the Japan bubbles. As expected, in a trending market where the prices
do not convert to fundamentals in a long time, investors will keep selling Japan ETFs
at low prices and buying it back at high prices, which hurts the performance of the
delta strategy. If we exclude the Japan from our investment universe, the performance
improves significantly: the gross return increase to 6.06% and Sharpe ratio to 0.38.
4. Robustness tests
4.1 Annually rebalance and GDP(PPP)-weighted portfolios
The benefit of rebalancing is well acknowledged, while the question of how often is
the portfolio to be rebalanced is still an open question. The above performance at
based on quarterly rebalance, in this section we also report the performance of
annually rebalanced portfolios. Since we do not have enough annual data points for
the optimal portfolios based on mean-variance analysis, only the performance of
simple portfolios is presented. We also consider an additional simple portfolio which
is weighted on the purchasing power parity GDP (GDP(PPP)). The purchasing power
parity takes into account the relative cost of living and the inflation rates of the
countries, rather than using just exchange rates which may distort the real differences
- 18 -
in income. We could not include the GDP(PPP)-weighted portfolio in the quarterly
rebalanced portfolio analysis because the GDP(PPP) data are only available at annual
frequency. Table 12 and table 13 report the annual performance of simple portfolios in
23 developed countries and 20 emerging markets.
The performance of annually rebalanced portfolios are very similar to that of quarterly
rebalanced portfolios and the main results of this paper are the same. The
equally-weighted portfolio achieves the highest Sharpe ratio and net returns in 23
developed countries. The GDP-weighted and GDP(PPP)-weighted portfolios have the
best Sharpe ratios and returns in 20 emerging markets, consistent to the results
discussed in section 3.1. The GDP(PPP)-weighted portfolio has better Sharpe ratio
and returns than the GDP-weighted portfolio, indicating additional benefit by taking
into consideration of inflation rates and relative cost of living.
4.2 Difference risk aversion levels
For all the asset allocation strategies considered above, we only report results for the
case in which risk aversion is equal to 2. We also consider the following levels of risk
aversion: γ = {1, 3, 5, 10}. The risk aversion levels only affect the performance of
mean-variance portfolio and the Bayes-Stein shrinkage portfolio. We find that the
results are not very different across risk aversion levels, only the CEQ return are
affected, because there is no optimization in the simple strategies, as risk aversion
increases the CEQ of the simple strategies drops.
5. Conclusion
This paper compares different optimal portfolio strategies with simple portfolio choices using
data from country index funds and ETFs and find how many counties are needed at least to
diversify globally.
We have the following contributions. We find the gains from optimal diversification is offset
by estimation errors, thus performance of the simple portfolios are very close to the optimal
portfolios based on mean-variance analysis in developed countries while outperform the
optimal portfolios in emerging markets.
Among the simple portfolio strategies, the most commonly used capitalization-weighted
- 19 -
portfolio suffers from price fluctuations and systematically over-weights the over-valued
countries and under-weights the under-valued countries. To overcome this shortcoming, we
propose the GDP weighting scheme based on the belief that the GDP is the most
straightforward fundamental at the country level and it is a relatively stable measure of the
size of each country in the global economy.
We further show the gains from diversification can be achieved in portfolios consisting of as
few as 3 developed countries or 4 emerging markets, in the case of portfolios consisting of
smaller number of countries, the estimation errors are less severe and thus, the optimal
portfolios have better out of sample performances.
Finally, we also find that a delta strategy based on the difference of a country’s GDP share and
capitalization share has a predictive power.
- 20 -
References
Arnott, R., J. Hsu, and P. Moore (2005). “Fundamental Indexation.” Financial
Analysts Journal, 61, 2, 83-99.
Brandt, M (2004). “Portfolio Choice Problem.” working paper.
DeMiguel , V., J. Nogales (2007). “Portfolio Selection with Robust Estimation.”
Working paper.
DeMiguel, V., L. Garlappi, R. Uppal (2007). “Optimal Versus Naive Diversification:
How Inefficient is the 1/N Portfolio Strategy?” Review of Financial Studies.
Estrada, J (2006). “Fundamental Indexation and International Diversification.”
Working paper.
Grubel, H (1968). “Internationally Diversified Portfolios: Welfare Gains and Capital
Flows.” American Economic Review, 58, 1299-1314.
Hamza, O., M, Kortas., J, LHer., M, Roberge (2006). “International Equity Portfolios:
Selecting the Right Benchmark for Emerging Markets.” Emerging Market Review, 7,
111-128.
Hsu, J (2006). “Cap-Weighted Portfolios are Sub-Optimal Portfolios.” Journal of
Investment Management, 4, 3, 1-10.
Hsu, J., and C. Campollo (2006). “New Frontiers in Index Investing. An Examination
of Fundamental Indexation.” Journal of Indexes, Jan/Feb, 32-58.
Jagannathan, R., T. Ma (2003). “Risk Reduction in Large Portfolios: Why Imposing
the Wrong Constraints Helps.” Journal of Finance 58(4) 1651–1684.
James, W., and C. Stein (1961). “Estimation with Quadratic Loss,” in Proceedings of
the 4th Berkeley Symposium on Probability and Statistics 1. Berkeley: University of
California Press.
Jorion, P (1985). “International Portfolio Diversification with Estimation Risk,”
Journal of Business, 58, 259–278.
Jorion, P (1986). “Bayes-Stein Estimation for Portfolio Analysis,” Journal of
Financial and Quantitative Analysis 21, 279-292.
Levy, H., M. Sarnat (1970). “International Diversification of Investment Portfolios.”
American Economic Review, 60, 4, 668-675.
Litterman, R. (2003). “Modern Investment Management: An Equilibrium Approach”.
Wiley.
Markowitz, H. M. (1952). “Mean-Variance Analysis in Portfolio Choice and Capital
Markets,” Journal of Finance, 7, 77–91.
- 21 -
Mayers, D. (1976). “Nonmarketable Assets, Market Segmentation, and the Level of
Asset Prices.” Journal of Financial and Quantitative Analysis, 11, 1, 1-12.
Roon, F. (2004). “On the Estimation Error in Mean-variance Efficient Portfolio
Weights.” Working paper.
Schoenfeld, S. (2006). “Are Alternatively Weighted Indexes Worth Their Weight?”
Northern Trust report, April.
Solnik, B. (1974). “Why Not Diversify Internationally Rather Than Domestically?”
Financial Analysts Journal, 30, 4, 48-54.
Stein, C. (1955). “Inadmissibility of the Usual Estimator for the Mean of a
Multivariate Normal Distribution,” in 3rd Berkely Symposium on Probability and
Statistics, , vol. 1, pp. 197–206, Berkeley. University of California Press.
Treynor, J. (2005). “Why Market-Valuation-Indifferent Indexing Works.” Financial
Analysts Journal, 61, 5, 65-69.
- 22 -
Table 1 Summary Statistics for 23 developed countries
Table 1 shows the monthly arithmetic mean return (Mean), standard deviation (STD),
risk-adjusted return (RAR=AM/STD), minimum (Min) and maximum (Max) return of each country
ETF over the Dec/1979-Dec/2006 period. It also shows the skewness and kurtosis of each country
ETF. All data are accounted for capital gains and dividends.
Country Mean STD RAR MIN MAX Skewness Kurtosis
Australia 1.22% 6.62% 18.43% -44.51% 20.79% -1.03 6.71
Austria 1.14% 6.51% 17.49% -23.29% 28.11% 0.45 2.37
Belgium 1.35% 5.70% 23.68% -18.85% 26.76% 0.23 3.11
Canada 1.02% 5.54% 18.49% -22.04% 17.98% -0.58 2.41
Denmark 1.37% 5.48% 24.99% -13.38% 21.30% 0 0.35
Finland 1.45% 9.33% 15.55% -31.76% 33.26% 0.13 1.24
France 1.21% 6.07% 19.86% -23.18% 21.03% -0.4 1.37
Germany 1.14% 6.36% 17.86% -24.35% 23.69% -0.27 1.64
Greece 1.80% 6.83% 26.28% -19.48% 21.04% -0.28 1.23
HK 1.48% 8.87% 16.74% -43.44% 33.23% -0.35 2.89
Ireland 1.11% 5.64% 19.74% -17.67% 18.42% -0.12 0.99
Italy 1.32% 7.18% 18.35% -20.39% 30.99% 0.3 0.84
Japan 0.96% 6.63% 14.53% -19.38% 24.26% 0.31 0.56
Netherlands 1.34% 5.22% 25.58% -17.80% 15.82% -0.53 1.44
New Zealand 0.82% 6.55% 12.49% -20.03% 27.65% 0.15 1.36
Norway 1.19% 7.36% 16.17% -27.85% 22.15% -0.44 1.24
Portugal 0.73% 6.38% 11.38% -19.36% 28.41% 0.31 1.6
Singapore 1.06% 7.52% 14.10% -41.34% 25.84% -0.54 4.03
Spain 1.45% 6.54% 22.16% -21.61% 26.72% 0.12 1.31
Sweden 1.72% 7.17% 24.07% -22.23% 22.91% -0.13 0.64
Switzerland 1.17% 5.16% 22.62% -17.64% 16.68% -0.15 1.1
UK 1.21% 5.26% 22.96% -21.53% 19.44% -0.1 1.57
US 1.13% 4.30% 26.41% -21.22% 13.28% -0.57 2.37
- 23 -
Table 2 Summary Statistics for 20 emerging markets
Table 2 shows the monthly arithmetic mean return (Mean), standard deviation (STD),
risk-adjusted return (RAR=AM/STD), minimum (Min) and maximum (Max) return of each country
ETF over the Dec/1994-Dec/2006 period. It also shows the skewness and kurtosis of each country
ETF All data are accounted for capital gains and dividends.
Country Mean STD RAR MIN MAX Skewness Kurtosis
Argentina 1.56% 11.59% 13.43% -31.36% 52.92% 0.33 2.64
Brazil 1.81% 11.45% 15.80% -37.63% 36.78% -0.39 1.38
Chile 0.77% 6.63% 11.59% -29.10% 20.13% -0.57 2.39
Colombia 1.74% 9.91% 17.52% -23.45% 30.61% 0.07 0.4
Czech 1.75% 8.19% 21.32% -27.57% 30.08% -0.15 1.43
Hungary 2.27% 10.28% 22.07% -38.80% 46.17% 0.1 3.36
India 1.18% 8.26% 14.28% -17.61% 22.13% -0.05 -0.54
Indonesia 1.26% 14.73% 8.55% -40.54% 55.59% 0.38 2.22
Israel 1.08% 7.36% 14.64% -18.85% 26.90% -0.12 0.97
Korea 1.32% 12.77% 10.32% -31.26% 70.60% 1.27 5.98
Malaysia 0.54% 9.84% 5.53% -30.20% 50.05% 0.83 6.02
Mexico 1.63% 8.69% 18.81% -34.25% 19.14% -0.83 1.64
Peru 1.53% 8.18% 18.68% -33.62% 36.11% -0.12 3.65
Philippines 0.00% 9.68% -0.02% -29.22% 43.39% 0.57 2.89
Poland 1.51% 10.58% 14.29% -34.82% 40.21% 0.27 1.56
Russia 3.48% 17.76% 19.57% -59.23% 61.13% 0.17 1.7
South Africa 1.08% 7.93% 13.59% -30.51% 19.45% -0.69 1.41
Taiwan 0.41% 8.79% 4.69% -21.73% 29.25% 0.4 0.66
Thailand 0.27% 12.76% 2.12% -34.01% 43.23% 0.37 1.75
Turkey 2.54% 16.89% 15.04% -41.24% 72.30% 0.57 2.1
- 24 -
Table 3. Portfolio performance in 23 developed countries quarterly rebalanced
Table 3 summarizes the quarterly performance of equally-weighted portfolio (EQUAL), the
GDP-weighted portfolio (GDP), the capitalization-weighted portfolio (CAP), the dividend-weighted
portfolio (DIV), composite portfolio (COMBO), global minimum variance portfolio(MIN),
mean-variance efficient portfolio (MVE), Bayes-Stein shrinkage portfolio (BS) and the MSCI world
index (MSCI_world) for 23 developed countries over the Dec/1979-Dec/2006 period. The
arithmetic mean of annualized gross return (Gross Return), standard deviation (STD) and beta
with respect to the MSCI world market index are reported in the table. Ex-post excess returns are
measure by Alpha, which is calculated using the CAPM model. Sharpe Ratio is the excess return
over risk-free rate divided by the standard deviation. The certainty-equivalent return (CEQ) is the
risk-free rate that an investor is willing to accept rather than adopting a risky portfolio strategy.
Turnover rate is the amount of trading required to implement each portfolio strategy. Assuming a
half percentage transaction cost per one-way transaction, the total Transaction Cost is calculated
as turnover*0.5%*2. Net return is Gross return net transaction cost.
Gross
Return STD Alpha Beta
Sharpe
Ratio CEQ Turnover
Transaction
Cost Net
Return
EQUAL 17.48% 15.40% 4.79%
0.90
0.75 15.11% 49.45% 0.49% 16.99%
GDP 14.39% 14.29% 1.15%
0.98
0.59 12.34% 41.37% 0.41% 13.97%
CAP 13.16% 14.86% -0.17%
0.99
0.49 10.95% 41.02% 0.41% 12.75%
DIV 14.58% 14.01% 1.49%
0.96
0.62 12.62% 36.71% 0.37% 14.22%
Combo 14.14% 14.17% 0.82%
0.99
0.58 12.13% 39.54% 0.40% 13.74%
MIN_exante 16.35% 13.91% 3.71%
0.90
0.73 14.41% 69.81% 0.70% 15.65%
MVE_exante 16.96% 21.97% 5.81%
0.70
0.50 12.13% 103.57% 1.04% 15.92%
BS 17.90% 16.65% 5.89%
0.81
0.72 15.13% 75.49% 0.75% 17.15%
MIN_expost 15.65% 12.62% 2.81%
0.92
0.77 14.06% 39.25% 0.39% 15.25%
MVE_expost 24.40% 17.31% 12.98%
0.73
1.07 21.41% 52.81% 0.53% 23.87%
MSCI_world 13.41% 14.34% 0.00%
1.00
0.52 11.35% 13.41%
- 25 -
Table 4. Portfolio performance in 22 developed countries excluding Japan quarterly
rebalanced
Table 4 summarizes the quarterly performance of equally-weighted portfolio (EQUAL), the
GDP-weighted portfolio (GDP), the capitalization-weighted portfolio (CAP), the dividend-weighted
portfolio (DIV), composite portfolio (COMBO), global minimum variance portfolio(MIN),
mean-variance efficient portfolio (MVE), Bayes-Stein shrinkage portfolio (BS) and the MSCI world
index (MSCI_world) for 22 developed countries excluding Japan over the Dec/1979-Dec/2006
period. The arithmetic mean of annualized gross return (Gross Return), standard deviation (STD)
and beta with respect to the MSCI world market index are reported in the table. Ex-post excess
returns are measure by Alpha, which is calculated using the CAPM model. Sharpe Ratio is the
excess return over risk-free rate divided by the standard deviation. The certainty-equivalent
return (CEQ) is the risk-free rate that an investor is willing to accept rather than adopting a risky
portfolio strategy. Turnover rate is the amount of trading required to implement each portfolio
strategies. Assuming a half percentage transaction cost per one-way transaction, the total
Transaction Cost is calculated as turnover*0.5%*2. Net return is Gross return net transaction
cost.
Gross
Return STD Alpha Beta
Sharpe
Ratio CEQ Turnover
Transaction
Cost Net
Return
EQUAL 17.75% 15.57% 5.19%
0.89
0.76 15.33% 49.03% 0.49% 17.26%
GDP 15.58% 14.49% 2.78%
0.92
0.67 13.48% 37.13% 0.37% 15.21%
CAP 14.80% 14.24% 2.03%
0.91
0.62 12.77% 33.50% 0.34% 14.46%
DIV 15.08% 14.27% 2.34%
0.91
0.64 13.04% 34.21% 0.34% 14.74%
Combo 15.24% 14.27% 2.44%
0.92
0.65 13.20% 35.01% 0.35% 14.89%
MIN_exante 16.44% 14.07% 3.93%
0.88
0.75 14.46% 67.79% 0.68% 15.77%
MVE_exante 16.59% 22.41% 5.64%
0.67
0.48 11.57% 103.08% 1.03% 15.56%
BS 17.98% 16.84% 6.03%
0.80
0.72 15.15% 74.00% 0.74% 17.24%
MIN_expost 16.13% 12.87% 3.53%
0.89
0.79 14.47% 37.36% 0.37% 15.76%
MVE_expost 24.41% 17.31% 12.99%
0.73
1.07 21.41% 52.80% 0.53% 23.88%
MSCI_world 13.41% 14.34% 0.00%
1.00
0.52 11.35% 13.41%
- 26 -
Table 5. Portfolio performance in 20 emerging markets quarterly rebalanced
Table 5 summarizes the quarterly performance of equally-weighted portfolio (EQUAL), the
GDP-weighted portfolio (GDP), the capitalization-weighted portfolio (CAP), the dividend-weighted
portfolio (DIV), composite portfolio (COMBO), global minimum variance portfolio(MIN),
mean-variance efficient portfolio (MVE), Bayes-Stein shrinkage portfolio (BS) and the MSCI
emerging market index (MSCI_emerging) for 20 emerging markets over the Dec/1994-Dec/2006
period. The arithmetic mean of annualized gross return (Gross Return), standard deviation (STD)
and beta with respect to the MSCI world market index are reported in the table. Ex-post excess
returns are measure by Alpha, which is calculated using the CAPM model. Sharpe Ratio is the
excess return over risk-free rate divided by the standard deviation. The certainty-equivalent
return (CEQ) is the risk-free rate that an investor is willing to accept rather than adopting a risky
portfolio strategy. Turnover rate is the amount of trading required to implement each portfolio
strategies. Assuming a half percentage transaction cost per one-way transaction, the total
Transaction Cost is calculated as turnover*0.5%*2. Net return is Gross return net transaction
cost.
Gross
Return STD Alpha Beta
Sharpe
Ratio CEQ Turnover
Transaction
Cost
Net
Return
EQUAL 25.11% 25.27% 7.78%
0.96
0.84 18.73% 85.34% 0.85% 24.26%
GDP 25.45% 24.19% 7.96%
0.97
0.89 19.60% 73.90% 0.74% 24.71%
CAP 20.99% 21.92% 3.34%
0.99
0.78 16.19% 69.25% 0.69% 20.30%
DIV 21.89% 22.46% 4.38%
0.98
0.80 16.84% 71.24% 0.71% 21.17%
Combo 23.04% 22.47% 5.45%
0.98
0.85 17.99% 71.61% 0.72% 22.33%
MIN_exante 19.48% 20.45% 3.37%
0.88
0.76 15.30% 95.17% 0.95% 18.53%
MVE_exante 19.66% 32.27% 4.55%
0.80
0.49 9.25% 95.77% 0.96% 18.70%
BS 23.80% 27.80% 6.58%
0.96
0.72 16.08% 84.44% 0.84% 22.96%
MIN_expost 22.65% 17.07% 6.29%
0.89
1.10 19.74% 66.71% 0.67% 21.99%
MVE_expost 33.87% 25.01% 18.79%
0.80
1.20 27.62% 93.05% 0.93% 32.94%
MSCI_emerging 17.84% 21.11% 0.00%
1.00
0.66 13.38% 17.84%
- 27 -
Table 6 Portfolio Weights in 23 developed countries
Table 6 shows the average weight of each country in the 10 portfolio strategies over the
Dec/1979-Dec/2006 period. The order of the countries is according to their average monthly
returns from high to low. Concentrate rate is the Herfindahl-Hirschman index.
Country
Monthly
return EQUAL GDP CAP DIV Combo MIN_exante MVE_exante BS MIN_expost MVE_expost
Greece 1.80% 4.35% 0.70% 0.33% 0.47% 0.50% 0.00% 0.00% 0.34% 0.00% 0.00%
Sweden 1.72% 4.35% 1.16% 0.74% 0.82% 0.90% 0.67% 20.91% 8.64% 0.00% 27.52%
HK 1.48% 4.35% 0.53% 1.52% 2.43% 1.49% 0.04% 21.98% 5.96% 0.00% 11.58%
Finland 1.45% 4.35% 0.53% 0.48% 0.54% 0.51% 1.33% 3.72% 1.49% 0.00% 0.00%
Spain 1.45% 4.35% 2.50% 1.31% 1.95% 1.80% 5.27% 0.00% 2.67% 0.00% 3.45%
Denmark 1.37% 4.35% 0.71% 0.30% 0.23% 0.35% 6.48% 7.31% 8.56% 6.72% 27.08%
Belgium 1.35% 4.35% 1.09% 0.56% 0.79% 0.82% 3.94% 0.84% 3.98% 3.86% 8.40%
Netherlands 1.34% 4.35% 1.66% 2.04% 3.19% 2.34% 3.43% 4.02% 4.12% 0.00% 0.00%
Italy 1.32% 4.35% 5.32% 1.74% 2.07% 3.06% 0.91% 2.84% 3.23% 0.56% 0.00%
Australia 1.22% 4.35% 1.80% 1.40% 2.46% 1.89% 4.38% 0.00% 4.28% 5.76% 0.00%
France 1.21% 4.35% 6.62% 2.98% 4.31% 4.63% 0.00% 0.00% 3.73% 0.00% 0.00%
UK 1.21% 4.35% 5.61% 8.83% 15.69% 10.01% 3.28% 0.00% 3.19% 4.68% 0.00%
Norway 1.19% 4.35% 0.69% 0.21% 0.24% 0.38% 0.00% 1.18% 5.35% 0.00% 0.00%
Switzerland 1.17% 4.35% 1.20% 1.77% 1.44% 1.47% 1.15% 3.63% 3.43% 5.17% 2.52%
Austria 1.14% 4.35% 0.85% 0.13% 0.11% 0.23% 12.79% 11.16% 9.96% 11.10% 19.44%
Germany 1.14% 4.35% 9.07% 3.76% 3.52% 5.49% 0.00% 0.00% 3.01% 0.00% 0.00%
US 1.13% 4.35% 37.42% 41.35% 46.50% 42.13% 37.47% 5.56% 5.11% 46.71% 0.00%
Ireland 1.11% 4.35% 0.43% 0.22% 0.28% 0.27% 0.04% 0.00% 1.79% 0.00% 0.00%
Singapore 1.06% 4.35% 0.27% 0.56% 0.54% 0.46% 1.51% 3.08% 6.83% 0.36% 0.00%
Canada 1.02% 4.35% 3.20% 2.46% 2.80% 2.84% 7.05% 0.00% 2.85% 0.00% 0.00%
Japan 0.96% 4.35% 17.95% 26.95% 8.99% 17.91% 5.61% 8.77% 2.81% 10.00% 0.00%
New_Zealand 0.82% 4.35% 0.24% 0.13% 0.33% 0.24% 1.93% 5.00% 7.66% 3.18% 0.00%
Portugal 0.73% 4.35% 0.47% 0.21% 0.30% 0.30% 2.73% 0.00% 1.04% 1.90% 0.00%
Concentration 4.35% 19.38% 25.61% 25.63% 22.84% 17.94% 12.94% 5.79% 25.62% 20.92%
- 28 -
Table 7 Portfolio Weights in 20 emerging markets
Table 7 shows the average weight of each country in the 10 portfolio strategies over the
Dec/1994-Dec/2006 period. The order of the countries is according to their average monthly
returns from high to low. Concentrate rate is the Herfindahl-Hirschman index.
Country
Monthly
return EQUAL GDP CAP DIV Combo MIN_exante MVE_exante BS MIN_expost MVE_expost
Argentina 1.56% 5.00% 5.95% 2.22% 3.81% 4.06% 0.41% 0.41% 4.96% 0.00% 0.00%
Brazil 1.81% 5.00% 15.98% 11.66% 20.00% 15.53% 0.15% 0.15% 4.22% 0.00% 0.00%
Chile 0.77% 5.00% 1.73% 3.60% 4.91% 3.40% 3.83% 3.83% 3.84% 9.90% 0.00%
Colombia 1.74% 5.00% 2.07% 0.70% 0.73% 1.17% 16.18% 16.18% 3.97% 8.47% 14.02%
Czech 1.75% 5.00% 1.43% 0.93% 1.54% 1.31% 16.31% 16.31% 4.43% 13.75% 23.09%
Hungary 2.27% 5.00% 1.20% 0.89% 0.86% 0.98% 0.00% 0.00% 10.47% 0.00% 28.37%
India 1.18% 5.00% 9.09% 8.18% 6.15% 7.68% 9.27% 9.27% 4.42% 9.32% 0.00%
Indonesia 1.26% 5.00% 3.90% 2.32% 2.12% 2.48% 0.00% 0.00% 3.00% 0.00% 0.00%
Israel 1.08% 5.00% 2.44% 2.53% 4.04% 3.05% 24.29% 24.29% 5.88% 27.50% 0.00%
Korea 1.32% 5.00% 11.45% 11.74% 8.52% 10.43% 1.62% 1.62% 4.16% 0.00% 0.00%
Malaysia 0.54% 5.00% 1.95% 6.94% 7.08% 6.29% 1.48% 1.48% 3.43% 4.66% 0.00%
Mexico 1.63% 5.00% 11.50% 7.97% 5.32% 8.37% 0.00% 0.00% 4.41% 0.00% 0.00%
Peru 1.53% 5.00% 1.28% 0.58% 0.87% 0.92% 5.13% 5.13% 4.72% 10.85% 5.68%
Philippines 0.00% 5.00% 1.74% 1.99% 1.11% 1.62% 2.58% 2.58% 2.95% 6.74% 0.00%
Poland 1.51% 5.00% 3.91% 1.58% 0.93% 2.14% 0.00% 0.00% 5.52% 0.00% 0.00%
Russia 3.48% 5.00% 7.06% 5.79% 3.73% 5.52% 0.00% 0.00% 10.27% 0.00% 27.70%
South Africa 1.08% 5.00% 2.94% 8.42% 10.98% 7.40% 16.74% 16.74% 4.06% 5.88% 0.00%
Taiwan 0.41% 5.00% 6.86% 15.87% 11.23% 11.18% 2.01% 2.01% 4.08% 2.94% 0.00%
Thailand 0.27% 5.00% 3.17% 2.89% 2.90% 2.96% 0.01% 0.01% 2.61% 0.00% 0.00%
Turkey 2.54% 5.00% 4.34% 3.20% 3.15% 3.52% 0.00% 0.00% 8.61% 0.00% 1.14%
Concentration � 5.00% 8.27% 8.66% 9.34% 8.05% 15.41% 15.41% 5.95% 14.30% 23.36%
- 29 -
Table 8. Portfolio performances in 3 developed countries quarterly rebalanced
Table 8 summarizes the quarterly performance of equally-weighted portfolio (EQUAL), the
GDP-weighted portfolio (GDP), the capitalization-weighted portfolio (CAP), the dividend-weighted
portfolio (DIV), composite portfolio (COMBO), global minimum variance portfolio(MIN),
mean-variance efficient portfolio (MVE), Bayes-Stein shrinkage portfolio (BS) and the MSCI world
index (MSCI_world) for 3 developed countries over the Dec/1979-Dec/2006 period. The
arithmetic mean of annualized gross return (Gross Return), standard deviation (STD) and beta
with respect to the MSCI world market index are reported in the table. Ex-post excess returns are
measure by Alpha, which is calculated using the CAPM model. Sharpe Ratio is the excess return
over risk-free rate divided by the standard deviation. The certainty-equivalent return (CEQ) is the
risk-free rate that an investor is willing to accept rather than adopting a risky portfolio strategy.
Turnover rate is the amount of trading required to implement each portfolio strategies. Assuming
a half percentage transaction cost per one-way transaction, the total Transaction Cost is
calculated as turnover*0.5%*2. Net return is Gross return net transaction cost.
Gross
Return STD Alpha Beta
Sharpe
Ratio CEQ Turnover
Transaction
Cost
Net
Return
EQUAL 13.64% 14.70% 0.54%
0.96
0.53 11.48% 38.80% 0.39% 13.25%
GDP 12.54% 14.28% -0.56%
0.96
0.46 10.50% 36.44% 0.36% 12.18%
CAP 12.09% 15.21% -1.08%
0.97
0.41 9.77% 38.53% 0.39% 11.70%
DIV 13.55% 13.81% 0.54%
0.95
0.55 11.64% 32.24% 0.32% 13.23%
Combo 12.78% 14.15% -0.43%
0.97
0.48 10.78% 35.71% 0.36% 12.43%
MIN_exante 13.79% 13.97% 0.67%
0.96
0.56 11.84% 35.86% 0.36% 13.43%
MVE_exante 14.55% 17.69% 2.06%
0.88
0.49 11.42% 54.61% 0.55% 14.00%
BS 13.63% 14.70% 0.53%
0.96
0.52 11.47% 38.94% 0.39% 13.24%
MIN_expost 13.81% 13.80% 0.72%
0.96
0.57 11.90% 32.96% 0.33% 13.48%
MVE_expost 14.87% 14.54% 2.31%
0.89
0.62 12.75% 30.59% 0.31% 14.56%
MSCI_world 13.41% 14.34% 0.00%
1.00
0.52 11.35% 13.41%
- 30 -
Table 9 Portfolio performances in 7 developed countries quarterly rebalanced
Table 9 summarizes the quarterly performance of equally-weighted portfolio (EQUAL), the
GDP-weighted portfolio (GDP), the capitalization-weighted portfolio (CAP), the dividend-weighted
portfolio (DIV), composite portfolio (COMBO), global minimum variance portfolio(MIN),
mean-variance efficient portfolio (MVE), Bayes-Stein shrinkage portfolio (BS) and the MSCI
emerging market index (MSCI_emerging) for 7 emerging markets over the Dec/1994-Dec/2006
period. The arithmetic mean of annualized gross return (Gross Return), standard deviation (STD)
and beta with respect to the MSCI world market index are reported in the table. Ex-post excess
returns are measure by Alpha, which is calculated using the CAPM model. Sharpe Ratio is the
excess return over risk-free rate divided by the standard deviation. The certainty-equivalent
return (CEQ) is the risk-free rate that an investor is willing to accept rather than adopting a risky
portfolio strategy. Turnover rate is the amount of trading required to implement each portfolio
strategies. Assuming a half percentage transaction cost per one-way transaction, the total
Transaction Cost is calculated as turnover*0.5%*2. Net return is Gross return net transaction
cost.
Gross
Return STD Alpha Beta
Sharpe
Ratio CEQ Turnover
Transaction
Cost
Net
Return
EQUAL 20.83% 21.14% 3.54%
0.96
0.80 16.36% 71.91% 0.72% 20.11%
GDP 24.95% 25.28% 11.32%
0.70
0.83 18.56% 67.75% 0.68% 24.27%
CAP 19.68% 22.11% 5.91%
0.71
0.71 14.79% 62.93% 0.63% 19.05%
DIV 21.93% 23.29% 8.34%
0.69
0.77 16.51% 64.97% 0.65% 21.28%
Combo 21.74% 22.73% 8.00%
0.71
0.78 16.57% 63.99% 0.64% 21.10%
MIN_exante 18.36% 18.14% 2.87%
0.83
0.80 15.07% 83.75% 0.84% 17.52%
MVE_exante 21.91% 22.89% 5.48%
0.90
0.79 16.67% 91.13% 0.91% 21.00%
BS 20.82% 21.13% 3.69%
0.95
0.80 16.35% 74.55% 0.75% 20.07%
MIN_expost 17.51% 17.83% 1.96%
0.84
0.76 14.34% 64.50% 0.64% 16.87%
MVE_expost 26.93% 23.87% 10.35%
0.91
0.96 21.23% 64.22% 0.64% 26.29%
MSCI_emerging 17.84% 21.11% 0.00%
1.00
0.66 13.38% 17.84%
- 31 -
Table 10 Portfolio performances in 4 developed countries quarterly rebalanced
Table 10 summarizes the quarterly performance of equally-weighted portfolio (EQUAL), the
GDP-weighted portfolio (GDP), the capitalization-weighted portfolio (CAP), the dividend-weighted
portfolio (DIV), composite portfolio (COMBO), global minimum variance portfolio(MIN),
mean-variance efficient portfolio (MVE), Bayes-Stein shrinkage portfolio (BS) and the MSCI
emerging market index (MSCI_emerging) for 7 emerging markets over the Dec/1994-Dec/2006
period. The arithmetic mean of annualized gross return (Gross Return), standard deviation (STD)
and beta with respect to the MSCI world market index are reported in the table. Ex-post excess
returns are measure by Alpha, which is calculated using the CAPM model. Sharpe Ratio is the
excess return over risk-free rate divided by the standard deviation. The certainty-equivalent
return (CEQ) is the risk-free rate that an investor is willing to accept rather than adopting a risky
portfolio strategy. Turnover rate is the amount of trading required to implement each portfolio
strategies. Assuming a half percentage transaction cost per one-way transaction, the total
Transaction Cost is calculated as turnover*0.5%*2. Net return is Gross return net transaction
cost.
Gross
Return STD Alpha Beta
Sharpe
Ratio CEQ Turnover
Transaction
Cost
Net
Return
EQUAL 26.38% 24.86% 9.88%
0.90
0.90 20.20% 73.31% 0.73% 25.65%
GDP 28.30% 26.74% 14.90%
0.68
0.91 21.15% 66.37% 0.66% 27.64%
CAP 26.02% 25.61% 12.50%
0.69
0.86 19.46% 63.01% 0.63% 25.39%
DIV 28.13% 28.65% 15.03%
0.66
0.84 19.92% 65.24% 0.65% 27.47%
Combo 26.81% 26.45% 13.43%
0.68
0.87 19.82% 64.05% 0.64% 26.17%
MIN_exante 23.61% 23.31% 8.69%
0.79
0.84 18.18% 78.69% 0.79% 22.82%
MVE_exante 21.39% 22.98% 5.05%
0.89
0.76 16.11% 89.55% 0.90% 20.49%
BS 26.38% 24.86% 9.85%
0.91
0.90 20.20% 75.31% 0.75% 25.62%
MIN_expost 24.25% 22.68% 8.64%
0.84
0.90 19.11% 63.06% 0.63% 23.62%
MVE_expost 27.50% 24.68% 10.95%
0.91
0.96 21.41% 63.89% 0.64% 26.86%
MSCI_emerging 17.84% 21.11% 0.00%
1.00
0.66 13.38% 17.84%
- 32 -
Table 11 Portfolio performances of Delta strategies
Table 11 summarizes the quarterly performance of the Delta strategy in 23 developed countries
over the Dec/1979-Dec/2006 period and 20 emerging markets over the Dec/1994-Dec/2006
period. The arithmetic mean of annualized gross return (Gross Return), standard deviation (STD)
and beta with respect to the MSCI world market index are reported in the table. Ex-post excess
returns are measure by Alpha, which is calculated using the CAPM model. Sharpe Ratio is the
gross return divided by the standard deviation. The certainty-equivalent return (CEQ) is the
risk-free rate that an investor is willing to accept rather than adopting a risky portfolio strategy.
Turnover rate is the amount of trading required to implement each portfolio strategies. Assuming
a half percentage transaction cost per one-way transaction, the total Transaction Cost is
calculated as turnover*0.5%*2. Net return is Gross return net transaction cost.
�
Gross
Return STD Alpha Beta
Sharpe
Ratio CEQ Turnover
Transaction
Cost
Net
Return
Develop 0.71% 14.66% -3.80%
(0.19)
(0.36) -1.44% 37.68% 0.38% 0.34%
Develop_exJapan 6.06% 16.05% 1.49%
(0.18)
0.38 3.48% 27.60% 0.28% 5.78%
Emerging 29.58% 22.85% 21.52%
0.18
1.12 24.36% 66.66% 0.67% 28.92%
- 33 -
Table 12. Portfolio performance in 23 developed countries annually rebalanced
Table 12 summarizes the annually performance of equally-weighted portfolio (EQUAL), the
GDP-weighted portfolio (GDP), the capitalization-weighted portfolio (CAP), the dividend-weighted
portfolio (DIV), composite portfolio (COMBO), global minimum variance portfolio(MIN),
mean-variance efficient portfolio (MVE), Bayes-Stein shrinkage portfolio (BS) and the MSCI world
index (MSCI_world) for 23 developed countries over the Dec/1979-Dec/2006 period. The
arithmetic mean of annualized gross return (Gross Return), standard deviation (STD) and beta
with respect to the MSCI world market index are reported in the table. Ex-post excess returns are
measure by Alpha, which is calculated using the CAPM model. Sharpe Ratio is the excess return
over risk-free rate divided by the standard deviation. The certainty-equivalent return (CEQ) is the
risk-free rate that an investor is willing to accept rather than adopting a risky portfolio strategy.
Turnover rate is the amount of trading required to implement each portfolio strategies. Assuming
a half percentage transaction cost per one-way transaction, the total Transaction Cost is
calculated as turnover*0.5%*2. Net return is Gross return net transaction cost.
�
Gross
Return STD Alpha Beta
Sharpe
Ratio CEQ Turnover
Transaction
Cost
Net
Return
EQUAL 17.32% 23.32% 4.56%
0.88
0.49 11.88% 36.03% 0.36% 16.96%
GDP 15.27% 19.77% 2.46%
0.89
0.47 11.36% 27.82% 0.28% 14.99%
PPP 16.05% 20.11% 3.28%
0.88
0.50 12.00% 27.38% 0.27% 15.78%
CAP 14.12% 18.73% 1.14%
0.91
0.44 10.61% 25.16% 0.25% 13.87%
DIV 14.85% 16.46% 2.25%
0.86
0.54 12.14% 22.60% 0.23% 14.63%
MSCI_world 13.41% 14.34% 0.00% 1.00 0.52 11.35% � � 13.41%
- 34 -
Table 13. Portfolio performance in 20 emerging markets annually rebalanced
Table 13 summarizes the annually performance of equally-weighted portfolio (EQUAL), the
GDP-weighted portfolio (GDP), the capitalization-weighted portfolio (CAP), the dividend-weighted
portfolio (DIV), composite portfolio (COMBO), global minimum variance portfolio(MIN),
mean-variance efficient portfolio (MVE), Bayes-Stein shrinkage portfolio (BS) and the MSCI
emerging market index (MSCI_emerging) for 20 emerging markets over the Dec/1994-Dec/2006
period. The arithmetic mean of annualized gross return (Gross Return), standard deviation (STD)
and beta with respect to the MSCI world market index are reported in the table. Ex-post excess
returns are measure by Alpha, which is calculated using the CAPM model. Sharpe Ratio is the
excess return over risk-free rate divided by the standard deviation. The certainty-equivalent
return (CEQ) is the risk-free rate that an investor is willing to accept rather than adopting a risky
portfolio strategy. Turnover rate is the amount of trading required to implement each portfolio
strategies. Assuming a half percentage transaction cost per one-way transaction, the total
Transaction Cost is calculated as turnover*0.5%*2. Net return is Gross return net transaction
cost.
�
Gross
Return STD Alpha Beta
Sharpe
Ratio CEQ Turnover
Transaction
Cost
Net
Return
EQUAL 25.27% 29.31% 3.33%
0.99
0.73 16.68% 69.09% 0.69% 24.58%
GDP 26.31% 33.96% 4.31%
0.99
0.66 14.78% 67.97% 0.68% 25.63%
PPP 28.66% 35.57% 6.71%
0.99
0.70 16.01% 69.12% 0.69% 27.97%
CAP 22.30% 30.19% 0.24%
1.00
0.61 13.18% 58.96% 0.59% 21.71%
DIV 21.86% 31.43% -0.04%
0.99
0.57 11.98% 57.22% 0.57% 21.29%
MSCI_emerging 17.84% 21.11% 0.00% 1.00 0.66 13.38% � � 17.84%
- 35 -
Figure 1 Efficient Frontiers for 23 developed countries
Figure 1 places the performance of equally-weighted portfolio (EQUAL), the GDP-weighted
portfolio (GDP), the capitalization-weighted portfolio (CAP), the dividend weighted portfolio (DIV),
composite portfolio (COMBO), global minimum variance portfolio(MIN), mean-variance efficient
portfolio (MVE), Bayes-Stein shrinkage portfolio (BS) and the MSCI world index (MSCI_world) for
23 developed countries over the Dec/1979-Dec/2006 period in the mean-variance space.
- 36 -
Figure 2 Efficient Frontiers for 20 emerging markets
Figure 2 places the performance of equally-weighted portfolio (EQUAL), the GDP-weighted
portfolio (GDP), the capitalization-weighted portfolio (CAP), the dividend weighted portfolio (DIV),
composite portfolio (COMBO), global minimum variance portfolio(MIN), mean-variance efficient
portfolio (MVE), Bayes-Stein shrinkage portfolio (BS) and the MSCI emerging market index
(MSCI_em) for 20 emerging markets over the Dec/1994-Dec/2006 period in the mean-variance
space.
- 37 -
Figure 3 Time series weights on Japan in GDP-weighted and capitalization-weighted portfolios
- 38 -
Figure 4 Portfolio variance against number of countries in 23 developed countries
Figure 4 ranks the countries based on their capitalization values, and the x-axis is read as: top 1
country in terms capitalization (which is the US), top 2 countries in terms of capitalization (which
are the US and Japan), top3 countries in terms of capitalization (the US, Japan and the UK). The
y-axis is the equally-weighted, capitalization-weighted and GDP-weighted portfolio risk (STD).
- 39 -
Figure 5 Portfolio variance against number of countries in 23 developed countries
Figure 5.1 ranks the countries based on their capitalization values, and the x-axis is read as: top 1
country in terms capitalization, top 2 countries in terms of capitalization etc. The y-axis is the
equally-weighted, capitalization-weighted and GDP-weighted portfolio risk (STD).
Figure 5.2 ranks the countries based on their GDP values, and the x-axis is read as: top 1 country
in terms GDP, top 2 countries in terms of GDP etc. The y-axis is the e equally-weighted,
capitalization-weighted and GDP-weighted portfolio risk (STD).
- 40 -
Figure 6 Delta strategy short sell side weight on Japan
Figure 6 shows the short sell side weight on Japan of the delta strategy over time.