International Asset Allocation for Incompletely-informed

Investors
∗
Yin-Feng Gau, Mingshu Hua and Wenlin Wu
____________________________________________________________________

ABSTRACT

In this study, we attempt to explain the home bias puzzle by examining the effect of information

quality on the asset allocation decisions taken by agents under the assumption that two types of

agents possess different information on the true expected returns of home and foreign assets. Our

calibration results based on MSCI data indicate that in order to hedge for the changing quality of

the information, when updating their estimates of foreign expected returns, those agents who are

partially-informed and relatively more conservative will tend to hold fewer foreign assets than

completely- informed agents. We find that with an increase in the precision of the estimates of

partially-informed agents, there is a corresponding reduction in the magnitude of their home bias.

Finally, with an increase in the instantaneous correlation between the returns of the home and

foreign assets, there is a similar corresponding decline in the magnitude of the home bias in the

portfolios of partially-informed agents.

Keywords: Asset allocation; Home bias; Incomplete information; Kalman-Bucy filter;

Learning.

JEL Classification: G11.

∗
Yin-Feng Gau is at the Department of Finance, National Central University, Taiwan; Mingshu Hua is at the
Department of Banking and Finance, Tamking University of Science and Technology, Taiwan; and Wenlin Wu
(the corresponding author) is at the Department of International Business Studies, National Chi Nan University,
Taiwan. Address for correspondence: Wenlin Wu, Department of International Business Studies, National Chi Nan
University, 1 University Road, Puli, Nantou County 545, Taiwan, ROC; Tel: 886-4-23861028; Fax:
886-4-22523456; e-mail: s0212904@ncnu.edu.tw.
0
1. INTRODUCTION
Portfolio theory suggests that in equilibrium, under the dual assumptions that all information

is freely available to all market participants and that there are no other barriers to portfolio

formation, all assets are held on the basis of their capitalization weights. Nevertheless, as

Table 1 clearly shows, home bias is readily apparent in allocations to international equities,

with investors tending to hold larger shares in home assets than their relative world market

capitalization size would suggest. That is, investors routinely violate conventional rules of

thumb, exposing themselves to more individual, country-specific risk (Lewis, 1999; Ahearne

et al., 2004).

<Table 1 is inserted about here>

According to Lewis (1999) and Karolyi and Stulz (2003), the potential explanations for the

home bias puzzle come under four categories: (i) institutional barriers, such as exchange rate

risk and transaction costs; (ii) asymmetric information; (iii) hedging motives; and (iv)

behavioral differences, whilst Sercu and Vanpee (2007) suggest that the home bias may actually

be caused by a mixture of rational decision making and bounded rationality.1 The most

convincing explanations are those relying upon asymmetric information and behavioral

differences; however, these individual factors cannot completely account for this effect.

Within the extant literature, there is a common assumption that when determining their

foreign investment strategies, investors with complete information are fully aware of the true

expected returns. However, the precise estimation of expected returns is far more difficult than

estimating the variance in returns; 2 thus, estimation errors are increasingly likely when

investors have to consider the expected returns of both home and foreign assets. This tendency

1
It is empirically demonstrated in a number of studies that home bias is derived from a mixture of
informational and behavioral explanations; see for example, Morse and Shive (2006), Karlsson and Norden
(2007) and Lutje and Menkhoff (2007).
2
Merton (1980) argues that any estimation of expected returns should include the effects of changes in the
level of risk associated with the market, whilst Lundtofte (2009) asserts that the expected excess return on any
asset is largely dependent upon its covariance not only with the market portfolio, but also with changes in the
estimates of the representative agent.

1
thereby creates a problem for the current explanations of the home equity bias (Baele et al.,

2007), as well as concerns as to whether the econometric assumptions of an estimation

procedure are consistent with the assumptions and structure of the original portfolio model

(Feldman, 2007).

When the true expected returns are unobservable by investors, their learning will come

from the realized means of the returns, upon which they can then base their investment

decisions on estimations; however, thus far, very few studies have examined the effects of

perceived expected returns on international portfolio choices. Despite the enormous changes on

a global scale in technology within the investment environment, all of which have contributed

to a reduction in information asymmetry, the phenomenon of home bias still persists, and

cannot be fully explained by an asymmetric information model (Sercu and Vanpee, 2007);

indeed, as noted by Ahearne et al. (2004) there has been only a slight decline in the level of

home bias over the past two decades.

In order to address these issues and to explain the evolution of investors’ portfolio

decision rules, this study proposes an alternative model which attempts to clarify the ways in

which changes in the estimates of unobservable expected returns may affect the foreign asset

allocations of investors. Our primary aim is therefore to attempt to explain the home bias

puzzle based upon information quality and the effects on the asset allocation decisions taken

by agents when the expected returns of the home and foreign assets are unobservable; in such

cases, investors are likely to use publicly available information to make their portfolio

choices.

We estimate the expected returns according to the Kalman-Bucy filter under a framework

of partial equilibrium, and then derive the partially-informed agents’ optimal portfolio weights

for a perceived global investment opportunity. By taking as a benchmark the optimal portfolio

weight obtained from a standard mean-variance optimization, we also attempt to explain the

2
home bias by determining the potential bias in the portfolio weights of partially-informed

investors.

Although it has been assumed in some studies (such as Alder and Dumas, 1983) that

investors react to unexpected changes in equilibrium in a ‘true’ economic environment, an

incomplete information model focuses instead on the ways in which investors respond to

unexpected changes in the ‘perceived’ economic environment.3 A detailed survey of the use of

the incomplete information model is provided by Feldman (2007),4 with many other studies

having used the model to examine the asset pricing effect of incomplete information on

economic growth within a partial or general equilibrium framework.5 However, a common

theme in all of these studies is their consideration of only a domestic model, as opposed to an

international model.

Asymmetric information and incomplete information models offer different explanations

of investors’ portfolio choices,6 with the former invariably focusing on the differences in

information content for assets across countries (Gehrig, 1993; Zhou, 1998). Thus, investors who

are risk averse will tend to concentrate on home assets, essentially because they have

insufficient information on the foreign assets. This hypothesis also posits that less informed

agents learn from more informed agents.

In contrast, the incomplete information model assumes that investors possess a

sub-filtration and prior beliefs, such that a change in their prior beliefs means a change in the

posterior beliefs, and thus, in the perceived expected returns within the asset price process

(Back, 2004). Since they are incompletely informed, the expected returns of these agents will

suffer from estimation errors. Partially-informed investors are also likely to behave more

3
Examples include Dothan and Feldman (1986), Detemple (1986, 1991) and Feldman (1989).
4
The complete information model is generally a special case of the incomplete information model. The same
methodology for solving for equilibria with incomplete information can be employed to solve for the case with
complete information, since it is possible to restate the incomplete information non-Markovian problem as an
σ-algebraic equivalent of the complete information Markovian problem (Feldman, 2007).
5
Examples include Brennan (1998), Riedel (2000), Feldman (2003) and Lundtofte (2006, 2008, 2009).
6
For a review, see Back (2004).

3
conservatively; and thus, they will tend to invest more in home assets which seem more familiar

to them. With more observations realized during the process of learning, investors will produce

more precise estimations of the unknown expected returns, resulting in reduced estimation errors

in our model analysis.

Information asymmetry plays an important role in the risk perceptions of investors

(Brennan and Cao, 1997), and this also holds for the incomplete information model. French

and Poterba (1991) demonstrate that investors have optimistic expectations relating to

domestic stock returns but pessimistic expectations of foreign stock returns.7 Kilka and

Weber (2000) suggest that these asymmetric expectations arise from investors’ perceptions of

knowledge or familiarity, including their own perceptions of their personal competence, or

subjective probability estimations.

As noted by both Huberman (2001) and Grinblatt and Keloharju (2001), the foreign

investment decisions of investors are largely related to their familiarity with home assets, and

since such ‘familiarity’ is invariably defined in terms of geographical proximity, there is

generally greater availability of information on domestic stocks, which is also relatively more

reliable. According to Massa and Simonov (2006), the real driving force behind portfolio

choices is information-based familiarity, which reflects a rational means of coping with

limited information.

Foreign assets seem more risky to investors who are unfamiliar with them, so these

investors may shy away from investing in such foreign assets, particularly in the emerging

markets, where information on fundamentals is neither reliable nor generally available to all

traders. Loose standards relating to financial statements and poor corporate governance result

7
Ziegler (2000) shows that in an incomplete information model, relative optimism is characterized by a
situation in which the market and individual investors produce quite diverse assessments of future dividend
growth; that is, investors have heterogeneous beliefs about future dividend growth. In contrast, we assume that
investors have different prior beliefs on the expected returns of home and foreign assets, essentially because of
their asymmetric information or their individual perceptions of competence. Feldman (2007) reviews the
persistence (or otherwise) of these heterogeneous beliefs.

4
in poor quality information on the related assets within these markets. Given such excessive

‘noise’ in the information on foreign assets, agents will place greater emphasis on their

subjective valuations.

Within our proposed model, such behavioral bias is incorporated into the updating

scheme of partially-informed agents relating to the expected returns of foreign assets.

According to Barberis (2000), the learning process may tilt the portfolio choice more, such

that the demand for hedging estimation risk may prevail over the demand for hedging

systematic uncertainty. Our model generalizes a standard incomplete information model by

considering a ‘behavioral’ extension which allows investors to follow their own idiosyncratic

investment rules.

Our calibration results, based on the MSCI indices (in US$), which cover the period

from January 1970 to December 2006, show that home bias occurs in the presence of

incomplete information and the conservatism of investors with regard to the expected returns

of foreign assets. Conservative agents appear to under-react to new information, adhering

instead to their own prior beliefs. We find that the quality of information, relative to the

precision of the estimates of the expected returns of investors, plays a key role in international

portfolio choices.

Three interesting features are highlighted in the present study. Firstly, the influence of the

prior beliefs of investors with regard to their optimal portfolio choices are largely dependent

upon their assessment of the estimation error. When investors have incomplete information on

domestic and foreign assets, there will be an increase (reduction) in their hedging demand for

the home asset with the variance in the prior on the expected returns of the domestic (foreign)

asset. However, their unfamiliarity with the foreign asset may result in such investors being

relatively conservative when updating their estimation of the mean returns of the foreign asset.

Thus, the magnitude of the hedging demand for the home asset is greater when investors

5
demonstrate greater conservatism towards foreign investment.

Secondly, even when investors possess lower quality initial information, they can

eventually produce a precise estimation of the expected returns through the dynamic learning

effect. Investors can improve the quality of their information by updating their beliefs based

upon more realized returns over time; over a sufficient period of time, their estimation errors

will ultimately disappear, and they can then adjust their portfolio choices to hold more foreign

assets. This thereby suggests that there is a general reduction in the level of home bias over

time, as found in both Ahearne et al. (2004) and Baele et al. (2007).

Thirdly, we derive a closed-form solution to the problem of international portfolio

allocation which allows for non-zero instantaneous correlation between the returns of the home

and foreign assets.8 Considering the linkages in the global markets, the sizeable flow of

information between them may relate to higher return correlations across these markets (Portes

and Rey, 2005). Within our model, prior expectations of the returns of the home and foreign

assets are linked by the correlation between them. As noted in both De Santis and Gerard (2006)

and Baele et al. (2007), the evident decline in the magnitude of home bias in Europe implies

that the establishment of the European Monetary Union (EMU) enhanced the transmission of

regional information, thereby reducing the overall level of home bias.

This study addresses the portfolio problem of Brennan (1998) and the outsider problem of

Lundtofte (2006), both of which assume a constant growth rate for an economy. We also extend

the prior one-country models to examine the international portfolio choices of

partially-informed investors, thereby providing an alternative explanation to the home bias

puzzle. Unlike Alder and Dumas (1983), who ignore the changing quality of information, we

demonstrate how partially-informed investors may hedge their choices according to the

changing quality or precision of the available information, and therefore tend to hold relatively

8
The ‘correlation’ within our model is defined as ‘instantaneous correlation’; however, in order to simplify our
explanation, we refer simply to ‘correlation’ throughout the remainder of the paper.

6
more home assets.

Furthermore, whilst Ueda (1999) assumes that no correlation exists between true and

observed local price processes – claiming that home bias is completely attributable to the

estimation errors of investors on the true process of foreign asset prices – we argue that home

bias may be the result of the unobservable expected returns of the home and foreign assets.

Furthermore, we consider that the updating mechanism of partially-informed agents may

relate not only to their prior beliefs, but also to their subjective under-reactions to new

information (i.e., conservatism) when updating their estimations of the expected returns of

foreign assets.

Finally, in contrast to Ueda (1999), we find that the portfolio choices of

partially-informed agents are related to their learning processes over time, as well as the

correlation between the returns of the home and foreign assets. We posit that investors can in

turn improve the quality of their information either through the learning effect or by obtaining

information from markets that are highly correlated.

The remainder of this paper is organized as follows. Section 2 outlines the problem of the

optimal portfolio choices of investors relating to international diversification, whilst also

providing the solution for optimal allocations to domestic equities by partially- and

fully-informed investors. Section 3 presents the numerical results and differences in the optimal

asset allocations of partially- and fully-informed agents. Finally, the conclusions drawn from

this study are presented in Section 4.

2. THE TWO-COUNTRY MODEL

2.1 The Economy

We adopt a two-country model in the present study, within which it is assumed that there two

types of agents, those who are fully-informed and those who are partially-informed, with each

of these agents possessing different information on the governing dynamics of the economy.

7
The economy contains three assets, a risk-free asset, a risky home equity and a risky foreign

equity. Fully-informed agents with filtration, F = {Ft}, have complete information on the true

diffusion processes of these three assets.

The risk-free asset in this case is a home-based money market account with price, Bt ,

which pays the locally constant risk-free rate, r; that is, dBt / Bt = rdt. Since we focus only on

the home equity bias, with the constant risk-free rate having no direct impact on this model, for

simplicity, we assume r = 0; therefore, Bt = 1 for all t.9 P1 and P2 are the respective prices of

the home and foreign equities, and both risky assets have a complete probability space (Ω, F, P)

which is a non-decreasing family of right-continuous sub-σ-algebraic elements, {Ft , 0 ≤ t ≤ T}.

The true diffusion processes of the home and foreign equities are therefore as follows:

dP1,t
= μ1dt + σ1dZ1 , (1)
P1,t

and

dP2,t
= μ2 dt + σ 2 ρ dZ1 + σ 2 1 − ρ 2 dZ2 , (2)
P2,t

where μ1 and μ2 are the respective constant expected returns of the home and foreign stocks

and σ1 and σ2 are the respective constant standard deviations of the home and foreign stocks.

Z1 and Z2 , which are independent standard Brownian motions, are defined on the complete

probability space (Ω, F, P) and dP1,t /P1,t and dP2,t /P2,t are correlated with the constant

instantaneous correlation, ρ, for which –1 < ρ < 1.

Since the partially-informed agents possess only limited information on the diffusion

processes relating to P1 and P2, they are therefore heavily dependent upon their prior

realizations of the home and foreign asset prices for their estimation of the unknown

9
This inference, which is similar to using the price of the money market account as the numeraire, has no effect on
the theoretical results. Other studies, such as Burger and Warnock (2007), also show that the home bias puzzle is
deepened when equity holdings are simultaneously considered with bonds.

8
parameters, as shown in Equations (1) and (2). Partially-informed investors have a filtration of
I I I
F = {Ft}, where Ft = σ(P1,s, P2,s; s ≤ t).

The assumption that μ1 and μ2 are unknown to these partially-informed agents creates

estimation errors for the expected returns of the home and foreign stocks, whereas the true

values of σ1, σ2 and ρ are known. This setting is consistent with the prior evidence showing

that the precise estimation of expected returns is far more difficult than the precise estimation

of variances (Merton, 1980; Lundtofte, 2009). Thus, the system of estimation for

partially-informed agents is dμi = 0, essentially because μi (i = 1, 2) is constant.10

Finally, following Alder and Dumas (1983) and Dothan and Feldman (1986), we assume

that the two types of agents maximize a time-additive, von Neumann- Morgenstern expected

utility of lifetime consumption.11 This utility function is strictly concave and increasing, as

well as being twice-differentiable. Each type of agent maximizes the expected utility of

lifetime consumption, conditional on the information set available at time t,

Max Et ⎡ ∫ V (Cs , s ) ds ⎤ ,
T
(3)
⎣⎢ t ⎦⎥

where C is the nominal rate of the agents’ consumption expenditure, and V(·) is a

homogeneous function within C of degree zero, which expresses the instantaneous rate of

10
The assumption of unobservable constant expected returns has significant implications, essentially because
the estimates by agents of the true expected returns are reduced to random walks, and thus, a source of instability.
Although Riedel (2000) shows that unobservable constant expected returns induce unbounded forward rates,
Feldman (2003) provides an explanation and resolution of this issue. Since we consider a situation in which the
true expected returns follow a stochastic process, the estimates of partially-informed agents, which track a
moving target, generally do not converge to the true parameter value with a Kalman filter (Brennan, 1998),
unless there is perfect correlation between the endowment flow and the expected returns, and unless a specific
technical condition is satisfied (Lundtofte, 2008). The detailed identification of the process paths of the
estimation errors and unobservable expected returns provided by Feldman (2007) leads to complete information.
Although home bias has been slightly reduced (Ahearne et al., 2004; Baele et al., 2007), analogous to the
portfolio problem examined in Brennan (1998) and the outsider problem dealt with in Lundtofte (2006), the
majority of the market participants, who are smaller investors, use only publicly available information to make
their portfolio choices. We assume that, over time, investors can learn of the true value of expected returns as
they observe more realized returns; thus, for expositional simplicity and comparison, we consider an
unobservable constant expected return on risky assets.
11
Although Detemple (1986) and Dothan and Feldman (1986), among others, used the von
Neumann-Morgenstern utility function, the more recent incomplete information models (including Brennan,
1998; Veronesi, 2000; and Lundtofte, 2006) assume that investors have a preference for constant relative risk
aversion (CRRA).

9
indirect utility. With the exception of the differences in the information structure, all agents

have identical preferences and wealth. All income sources other than asset returns are

excluded from the model.

2.2 The Filtering Problems of Partially-informed Agents

The partially-informed agents use publicly available information on the home and foreign

assets to estimate the expected returns μi (i = 1, 2). To restate the portfolio optimization

problem of the agents as a Markovian structure, we require the conditional distribution of the

unobservable μi (Feldman, 2007).

I 2 I
Let mi,t = E(μi| Ft ) and vi,t = E[(μi – mi,t ) | Ft ] be the respective conditional mean and
I
variance of μi, and let v12,t = v21,t = E[(μ1 – m1,t )(μ2 – m2,t )| Ft] be the conditional covariance of μ1

and μ2 , where vi,t (i = 1, 2); v12,t and v21,t are referred to as the filtering errors.12 Thus we have:

⎛ dP1,t ⎞
⎜ ⎟ σ
⎟ = ⎛ m1,t ⎞ dt + ⎜⎛ 1 ⎞ ⎛ dZ1 ⎞
0
⎜ P1,t ⎟ , (4)
⎜ dP2,t ⎜
⎟ m2,t ⎟ ⎜ σ ρ σ 1 − ρ 2 ⎟ ⎝⎜ dZ 2 ⎠⎟
⎜ ⎟ ⎝ ⎠ ⎝ 2 2 ⎠
⎜ P ⎟
⎝ 2,t ⎠

⎛ ⎛ dP1,t ⎞ ⎞
−1 ⎜ ⎜ ⎟ ⎟ −1
⎛ dZ1 ⎞ ⎛ σ 1 0 ⎞ ⎜ ⎜ P1,t ⎟ ⎛ m1,t ⎞ ⎟ ⎛ dZ1 ⎞ ⎛ σ1 0 ⎞ ⎛ μ1 − m1,t ⎞
where ⎜ ⎟ = ⎜⎜ ⎟ ⎜⎜ −
⎟ ⎜ ⎟ dt ⎟ = ⎜ ⎟ +⎜ ⎟ ⎜ ⎟ dt.
⎝ dZ2 ⎠ ⎝ σ 2 ρ σ 2 1 − ρ ⎟⎠ ⎜ ⎜ dP2,t ⎟ ⎝ m2,t ⎠ ⎟ ⎝ dZ2 ⎠ ⎝⎜σ 2 ρ σ 2 1− ρ ⎠⎟ ⎝ μ2 − m2,t ⎠
2 2

⎜ ⎜ P2,t ⎟ ⎟
⎝⎝ ⎠ ⎠
−1
⎛ σ1 0 ⎞ ⎛ μ1 − m1,t ⎞
The term ⎜⎜ ⎟
2 ⎟ ⎜ ⎟ is the market price of risk, which reflects a
⎝ σ 2 ρ σ 2 1 − ρ ⎠ ⎝ μ2 − m2,t ⎠

change from risk-neutral to real-world measures. Since the perceived gains or losses on the

asset returns of partially-informed agents differ from those of the true expected returns, they

will therefore undertake an upward or downward revision of their estimation of μi. According

12
In our model, the conditional means, which also minimize the mean squared errors, are incidental, essentially
because the validity of the estimates (i.e., the conditional moments) originates from agents’ spanning
sigma-algebras, which are equivalent to the original values (see Feldman, 2007). We are grateful to the referee
for raising this point.

10
 – –
to standard filtering theory and Liptser and Shiryaev (2001), Z 1 and Z 2 are independent
I
Brownian motions, conditional on the information set of the partially-informed agents, Ft.

This change in the measure does not alter the correlation between home and foreign asset

returns ( ρ), because the correlation is not affected by changes in the drift terms.

The partially-informed agents can improve their priors and estimates through the non-zero

correlation between home and foreign asset returns ( ρ), which links their prior beliefs on the

returns of the home and foreign assets; however, where the priors on the home and foreign asset

returns are correlated, it is difficult to obtain comparative static results. In order to simplify our

analysis, we assume that the agents’ estimation of μ1 is independent of their estimation of μ2 ,

conditional on the correlation between the asset returns, which implies that v12,t and v21,t are

zero for the partially-informed agents.

Next, let the agents have a Gaussian prior distribution with mean mi,0 and variance vi,0 at
I
time 0. The updating rules for the conditional means, mi,t = E(μi| Ft ), i = 1, 2, are:

⎛ v1,t − ρ v1,t ⎞
⎜ ⎟
⎛ dm1,t ⎞ ⎜ σ 1 σ 1 1 − ρ 2 ⎟ ⎛ dZ1 ⎞
⎟=⎜ (5)
⎜ ⎟ ⎜ dZ ⎟ ,
⎝ dm2,t ⎠ ⎜ 0 v2,t
⎟⎝ 2 ⎠
⎜ σ 2 1− ρ ⎠2 ⎟
⎝

We can also write the conditional variance-covariance matrix of μ1 and μ2 as:

⎛ v1,0σ 12 (1 − ρ 2 ) ⎞
⎜ 0 ⎟
⎛ v1,t 0 ⎞ ⎜ tv1,0 + σ 12 (1 − ρ 2 ) ⎟ (6)
⎜ ⎟= ⎟.
⎝ 0 v2,t ⎠ ⎜ v2,0σ 22 (1 − ρ 2 ) ⎟
⎜ 0
⎜ tv2,0 + σ 22 (1 − ρ 2 ) ⎟⎠
⎝

where vi,t , the filtering error relating to the expected return, μi , describes the uncertainty in the μi
I
parameter. Furthermore, the posterior distribution of μi follows normal distribution, {μi| Ft}~

11
N(mi,t , vi,t).13

As revealed by Equation (5), the filtering errors, vi,t , the volatility levels of the asset

returns, σi , and the correlation between home and foreign asset returns, ρ, will all affect the

change in the estimation of the expected returns, dmi,t. The magnitude of dmi,t becomes greater

in those cases where the asset returns are less volatile or where there are larger filtering errors.

Equation (5) also demonstrates that vi,t could be a measure of the quality of the agents’

information relating to μi . When vi,t = 0, for i = 1, 2, mi,t = μi , this indicates that the

partially-informed agents have obtained complete information relating to μi , such that they

can then produce a more precise estimation of the true expected returns.

Equation (6) indicates that the filtering error vi,t increases in the prior variance, vi,0 , and

declines to 0 as t goes to infinity. Equation (6) also shows that under the assumption of perfect

correlation between the returns of the home and foreign assets, or | ρ| = 1, then vi,t = 0, for i = 1,

2. This implies that no errors are exhibited in the agents’ estimation of the true expected

returns (μi , i = 1, 2) since the returns on both the home and foreign equities are perfectly

correlated. Equations (1) and (2) show that when | ρ| = 1, only one standard Brownian motion

exists within the economy, Z1. Under this scenario, there is only one risky asset in the model;

that is, the price processes of both the home and foreign equities are identical.

Given the initial condition, the filter represents the collection of moments in the

probability distribution of the unobservable variable, conditional on the available information;

however, the construction of the agents’ beliefs may be inferred based upon the optimal use of

the data (Berrada, 2009); in particular, the decision processes of the agents may be affected by

psychological irrationalities.14 Agents often exhibit optimistic expectations with regard to the

expected returns of home assets and pessimistic expectations on the expected returns of

13
The relevant proofs of Equations (5) and (6) are provided in the Appendix.
14
For example, Daniel et al. (1999) and Barberis et al. (2001) consider departures from rationality in the
representative-agent models to address important puzzles in asset pricing relating to the equity premium and the
risk-free interest rate

12
foreign investment (French and Poterba, 1991), essentially because tend to they impose their

subjective measures on the expected returns (Kilka and Weber, 2000). Research also suggests

that the extent of the relative optimism across assets can be defined as the difference between

return expectations across such assets (Shiller et al., 1996). This optimism, in turn, translates

into greater investment in more familiar assets.

In addition to mixing informational explanations with bounded rationality, some of the

more recent studies (such as Massa and Simonov, 2005) combine the incomplete information

model with the bounded rationality of investors to determine the effects of their subjective

measures of the mean estimates. In Berrada (2009), a coefficient is added into the learning

processes of investors so as to address their over- or under-reactions to new information. In

order to consider the possible bias in the learning mechanisms of investors on the mean

returns of the foreign assets, m2,t , we rewrite Equation (5) as a general formulation of

updating rules which nests the Bayesian learning as a special case.

⎛ v1,t − ρ v1,t ⎞
⎜ ⎟
⎛ dm1,t ⎞ ⎛ 1 0 ⎞ ⎜ σ1 σ 1 1 − ρ ⎟ ⎛ dZ1 ⎞
2
(7)
⎜ ⎟=⎜ ⎟⎜ ⎟ ⎜ dZ ⎟ ,
⎝ dm2,t ⎠ ⎝ 0 1 − ξ ⎠ ⎜ 0 v2,t
⎟⎝ 2 ⎠
⎜ 2 ⎟
σ 2 1− ρ ⎠
⎝

where ξ reflects the behavior of conservative partially-informed agents when updating their

expectations on the mean returns of the foreign asset, and 0 ≤ ξ ≤ 1.

When ξ = 0, there is no behavioral bias in the updating of m2,t by investors; Equation (7)

is equivalent to Equation (5), with classical Bayesian updating being represented by a

continuous time version of the Kalman filter. When 0 < ξ < 1, the partially-informed agents

exhibit conservatism with regard to updating their current estimate of the mean return, m2,t .

With an increase in ξ, the investors act more conservatively in response to new information in

their learning process of μ2 . The depression in the change in estimate m2,t is related to the

under-reaction by investors to new information; and indeed, when ξ = 1, the investors do not

13
react to any news or information which has recently become available.

Under this simple parameterization, Equation (7) can cover various types of behavior and

reactions, which thereby allows agents to have different learning rules; we can therefore

capture their learning bias in their updating of the estimates of the expected returns of the

foreign asset. In contrast to Equation (5), which provides an objective reference point for the

updating of the estimates of expected returns, in Equation (7), the bounded rationality of

investors is incorporated into the updating rules for the estimation of μ2 .

During the process of updating, conservative agents will tend to place greater weight on

their prior beliefs and less weight on new incoming information. This conservatism coefficient

can also be linked to the confidence amongst investors based upon familiarity, which thereby

suggests that the expectations of such investors relating to the risks and returns of unfamiliar

assets will tend to reveal systematic bias (Kang and Stulz, 1997; Huberman, 2001).

With 0 < ξ < 1, when updating m2,t , the impact of shocks will be devalued; this is

attributable to the conservatism amongst agents in their updating of the mean returns of

foreign equities with which they are unfamiliar. Given their unfamiliar with such foreign

assets, conservative agents under-react to new information when updating their estimates of

μ2 . This conservatism (or under-reaction mechanism) relating to updating may confirm one of

the explanations of the home bias, which is that investors’ expectations of future returns from

their foreign assets are consistently lower than the value predicted by a theoretical model.15

As noted in Massa and Simonov (2006), with an increase in the precision of information,

the home bias arising from learning bias is reduced in the updating rules, although parameter ξ

within our model can capture the behavioral bias of agents when forming their subjective

updating rules on the expected returns of the foreign assets. If investors have complete

15
Hasan and Simaan (2000) also note that lower expected returns are associated with estimation errors in the
mean vector of returns. Brennan (1998) demonstrates how estimation risks, which refers to the risk of learning
bad news, affect the incentives of agents to engage in equity investment. Such learning of bad news is related to
lower expected equity returns.

14
information on the foreign markets, their subjective bias in the learning rules, relative to the true

expected returns, should disappear.

2.3 Optimal Portfolio Choices

2.3.1 Portfolio choices of fully-informed agents

Given that the fully-informed agents are fully aware of the true expected returns of both the

foreign and home assets, their optimal portfolio decision can be solved by the following

maximization problem, conditional on the information set available at time t, or Ft :

J (WA , t ) = Max Et ⎡ ∫ V (C A, s , s)ds Ft ⎤ ,

T
(8)
C A ,s ,θ A ,s ⎣⎢ t ⎥⎦

{ }
s.t . dW A,t = ⎡⎣θ A,t μ1 + (1 − θ A,t ) μ 2 ⎤⎦ W A,t − C A,t dt + θ A,tW A,tσ 1dZ1

+ (1 − θ A,t )WA,tσ 2 ρ dZ1 + (1 − θ A,t )WA,tσ 2 1 − ρ 2 dZ 2

where θA,t is the ratio of the wealth invested in the home asset by the fully-informed agents,

and the equation for dWA,t is their dynamic budget constraint.16

When the market clears, at which time the problem in Equation (8) can be optimally

solved, this results in a situation of equilibrium. The optimal allocation to the home equity,

obtained from the first-order condition with respect to θA,t , is:

μ1 − μ 2 σ 22 − σ 1σ 2 ρ
θ A ,t = + , (9)
γ A (σ 12 + σ 22 − 2σ 1σ 2 ρ ) σ 12 + σ 22 − 2σ 1σ 2 ρ

2 2
where (σ 1 + σ 2 – 2σ1 σ2 ρ ) is the variance in the international portfolio, and γA refers to the

relative risk aversion of the fully-informed agents, derived from the inverse of their risk
–1
tolerance (– JWA / JWAWAWA ) .

The optimal allocation to the home asset by the fully-informed agents in Equation (9) is

similar to the weight applied in Lewis (1999). Such agents fairly allocate their wealth on the

basis of mean-variance optimization, without any bias across assets. The first term on the

16
A similar specification is adopted in Lewis (1999).

15
 μ1 − μ 2
left-hand side of Equation (9), , which relates to the maximization of
γ A (σ + σ 22 − 2σ 1σ 2 ρ )
2
1

portfolio expected returns, reveals the relative attraction of the assets.

Consistent with many of the studies on home bias, we consider the case of risk-averse

investors, for whom γA > 1. A higher value of γA indicates that investors require higher expected

returns for a given level of risk. When γA approaches infinity, θA,t is restated as
σ 22 − σ 1σ 2 ρ
, which suggests that fully-informed agents will allocate substantially more
σ 12 + σ 22 − 2σ 1σ 2 ρ
to the home asset when the foreign asset is riskier, and vice versa. Overall, Equation (9)

shows that moderately risk-averse investors are more concerned with the relative volatility

levels across assets than with the maximum expected returns of their portfolio; the

fully-informed agents’ optimal portfolio weights remain constant over time in this model.

2.3.2 Portfolio choices of partially-informed agents

When making their optimal portfolio choices, partially-informed agents rely on their prior

beliefs and publicly available information to estimate the true expected returns. Given the

learning bias in updating their beliefs relating to μ2 , our model considers a general form of

updating equation for m2,t , the agents’ beliefs at time t on the current value of the expected

returns, subject to their bounded rationality. The agents’ optimal portfolio decision can be

formulated as a Markovian problem (Feldman, 2007):

J (WB , m1 , m2 , t ) = Max Et ⎡ ∫ V (CB , s , m1,s , m2, s , s )ds Ft I ⎤ ,

T
(10)
CB ,s ,θ B ,s ⎣⎢ t ⎥⎦

{ }
s.t. dWB,t = ⎡⎣θB,t m1,t + (1 − θB,t )m2,t ⎤⎦ WB,t − CB,t dt + θB,tWB,tσ1dZ1

+(1 − θB,t )WB,tσ 2 ρ dZ1 + (1 − θB,t )WB,tσ 2 1 − ρ 2 dZ2 ,

v1,t ρ v1,t
dm1,t = dZ1 − dZ 2 , and
σ1 σ1 1 − ρ 2

16
 (1 − ξ )v2,t
dm2,t = dZ 2 ,
σ 2 1− ρ 2

where θB,t denotes the weight allocated to the home equity by the partially-informed agents,

and the equation for dWB,t is their dynamic budget constraint.

When the market clears, at which time Equation (10) can be optimally solved, the

optimal allocation to the home equity θB,t by the partially-informed agents is:

m1,t − m2,t σ 22 − σ1σ 2 ρ

θ B ,t = +
γ B (σ12 + σ 22 − 2σ1σ 2 ρ ) σ12 + σ 22 − 2σ1σ 2 ρ

⎛ γ −1 ⎞ ⎛ σ 12 ⎞ ⎛ v1,0 (1 − ρ 2 ) ⎞
+⎜ B ⎟⎜ 2 ⎟ ⎜⎜ 2 ⎟
(11)
⎟
⎝ γ B ⎠ ⎝ σ 1 + σ 2 − 2σ 1σ 2 ρ ⎠ ⎝ tv1,0 + σ 1 (1 − ρ ) ⎠
2 2

⎛ γ B − 1 ⎞⎛ σ 22 ⎞ ⎛ v2,0 (1 − ρ 2 ) ⎞
−(1 − ξ ) ⎜ ⎟⎜ 2 ⎟ ⎜⎜ 2 ⎟⎟
⎝ γ B ⎠⎝ σ1 + σ 2 − 2σ1σ 2 ρ ⎠ ⎝ tv2,0 + σ 2 (1 − ρ ) ⎠
2 2

where γB = – JWBWBWB / JWB denotes the relative risk aversion of partially-informed agents.

The first two terms in Equation (11), which are analogous to the optimal portfolio

allocation to the home equity by fully-informed agents, also form the myopic demand for the

home equity by partially-informed agents, whilst the last two terms indicate the hedging

demand amongst agents for the home equity. This hedging demand arises as a result of learning,

which represents the direct effect of the estimation errors in μ1 and μ2 ; the hedging demand

disappears only when there are no concurrent estimation errors in μi (i = 1, 2). Equation (11)

shows that the dynamic effect of learning relating to μi introduces the hedging demand

associated with the stochastic variation in μi . Several implications of this hedging demand

then follow.

Firstly, there is an increase (decrease) in this hedging demand with the variance in the

priors on the expected returns of the home asset, v1, 0 (foreign asset, v2, 0), which indicates that

the partially-informed agents will tend to invest more in the home asset when the estimation

17
error in the expected returns of the home (foreign) asset is higher (lower).

Secondly, this hedging demand for the domestic equity decreases with time horizon, t ,

which is the point at which the agents determine their asset allocation. This suggests that, over

time, their reduced holdings of the home equity are likely to be due to the learning effect, with

the partially-informed agents updating their estimates on the true mean returns upon the arrival

of more new information.

Thirdly, the hedging demand based upon the estimation errors in μ1 and μ2 is also

dependent upon the magnitude of the conservatism parameter, ξ, with the hedging demand

decreasing in (1 – ξ).17 When 0 < (1 – ξ) < 1, conservative agents under-react to new information

when updating their estimates on the expected returns of the foreign asset, with this

conservatism (or under-reaction) inducing greater hedging demand for the home asset. This

therefore implies that, with 0 < ξ < 1, conservative investors depress their updating of the

estimation of the mean returns of the foreign asset, regardless of whether there is good or bad

news. Ceteris paribus, the magnitude of the hedging demand is greatest for the home asset

with ξ = 1. Therefore, such subjective conservatism may well explain why partially-informed

agents may decide to place relatively more of their wealth in the home asset.

According to Equation (11), the sign and magnitude of the last two terms contributing to

the optimal portfolio composition of the agents is dependent upon the willingness of investors

to trade off the risks against the expected returns; that is, the degree of relative risk aversion

( γB). When γB > 1, the dominance of the substitution effect on the hedging demand leads to

partially-informed agents demanding more of the home asset in order to protect against the

changing quality of information. When γB → ∞, the demand amongst the agents for the home

equity is dependent upon the minimum variance portfolio share, determined by the second

term in Equation (11) and the hedging demand, derived from the third and fourth terms.

17
Irrespective of the value taken by ξ, it is still subject to the restriction of 0 ≤ θB ≤ 1.

18
 Alternatively, when 0 < γB < 1, the dominance of the income effects on the hedging

demand induces agents to hold less of the home asset. For γB = 1, this unit-relative risk

aversion is a knife-edge value, with switches occurring between the dominance of the

substitution and income effects. A balanced situation therefore emerges within our model; that

is, the hedging demand arising from the substitution effect may offset the demand attributable

to the income effect, resulting in zero hedging demand. As a result, partially-informed agents

will then behave like fully-informed agents.18

The level of risk in the returns of the home and foreign assets, respectively denoted as σ1

and σ2 , crucially affects the hedging demand for the home asset. If the foreign returns are

more volatile than the relative home returns, such that σ1 < σ2 , there will be an increase in

foreign investment risk, and the agents then have a higher hedging demand for the home asset,

but a smaller hedging demand for the foreign asset. Consequently, a home bias is discernible.

On the other hand, when σ1 > σ2 , we observe a reduction in the level of the home bias.

The correlation between the home and foreign asset returns ( ρ) therefore affects the value of
2 2
the agents’ international portfolio risk (σ 1 + σ 2 – 2σ1 σ2 ρ) and estimation errors (vi,t , i = 1, 2).

However, such a situation may well complicate the agents’ optimal portfolio strategy.

In an attempt to illustrate this effect, we focus on the case of 0 < ρ< 1; if the correlation
2
between asset returns is moderately high, the agents’ reduced international portfolio risk (σ1 +
2
σ2 – 2σ1 σ2 ρ) leads to an increase in their optimal allocation to the home equity, which erodes
the diversification benefit of holding foreign assets.19 However, this effect may actually be

more related to the better quality of information flows between the home and foreign markets

essentially because of the significant correlation between the home and foreign equity returns.

Based upon the enhanced availability of information transmission across the markets, the

18
We are extremely grateful to the referee for bringing this point to our attention.
19
Although the diversification benefits of foreign investment are weakened with an increase in market
correlation, they are still significant for investors (Heston and Rouwenhorst, 1994; Griffin and Karolyi, 1998;
Baele and Inghelbrecht, 2006).

19
agents may update their priors, leading to the production of more precise estimates of the

expected returns of both the home and foreign assets; as a result, the partially-informed agents

may no longer favor the home asset. According to these effects, the net influence of the

correlation between asset returns on θB,t cannot be determined, but as demonstrated by Portes

and Rey (2005), such agents become more concerned with the informative effect on

international portfolio decisions rather than with the motivation for diversification.

3. CALIBRATION RESULTS
We can gauge the magnitude of the home bias, in terms of the optimal portfolio choices, by

calibrating the incomplete information model on the returns from MSCI indices, including the

United States Index (US), the World Index ex USA Index (WI-ex-US), and the Emerging

Markets Index (EM), each of which is expressed in US dollars. The data on the monthly

returns of the US and WI-ex-US indices, which cover the period from January 1970 to

December 2006, feature a total of 444 observations. The data on the EM index are available

only from January 1988 to December 2006, thereby providing a sample of 228 observations.

Adopting the perspective of US agents, we treat the WI-ex-US and EM indices as

‘foreign’ assets; in particular, the WI-ex-US index provides a proxy for assets in the

developed markets. Since the developed markets (including the US) have a long history of

market development and stock return data, US investors benefit from the higher quality

information on the US and other developed markets; thus, they may tend to produce better

estimates of the expected returns on the US and WI-ex-US indices. US agents also enjoy an

informational advantage and are more familiar with the US market.

On the other hand, the lack of any long history or reliable stock returns data within the

emerging markets leads to US investors possessing only a limited amount of lower quality

information on the unfamiliar emerging markets. We therefore expect to find that the prior

beliefs of the US agents on the equity returns in the emerging markets will differ markedly

20
from their knowledge on the equity returns of the developed markets.

Table 2 presents a statistical summary of the monthly MSCI index returns, which shows

that over the sample period, the US was the least volatile market, with the lowest average

monthly returns, whilst the EM index was the most volatile, with the highest average returns;

ceteris paribus, the Sharpe ratios would suggest that the WI-ex-US index should have greater

attraction to US investors. All return series exhibit negative skewness, with the significantly

negative skewness of the EM index returns possibly being attributable to the financial crises

which occurred in the emerging markets between January 1988 and December 2006.

<Table 2 is inserted about here>

The correlation statistics in Panel B of Table 2 indicate the benefits to be gained from

diversification for US investors, through their investment in the foreign equity markets. The

correlation statistic between the US index and the EM index is 0.5973, whilst that between the

US index and the WI-ex-US index is 0.5775. The parameters of the means, standard

deviations and correlations are calibrated to the sample estimates based upon the whole

sample from 1970 to 2006 for the US and WI-ex-US indices, and from 1988 to 2006 for the

EM index. We assume that the fully-informed US agents are fully aware of the true values of

all of the parameters, whereas the partially-informed US agents are only aware of the true

standard deviations and correlations. The optimal allocations to the home asset are reported

for the trading period from January 1996 to December 2006.

The partially-informed agents use observations prior to the start date of the trading

period in order to form their priors. Provided that the number of observations in the estimation

period, K, the variance of agents’ common prior, is taken to be the variance in the estimate,
2
then vi,0 = σi /K. The value of K proxies for the precision of the priors. From Equation (6), we
2 2
σi (1 – ρ )
obtain vi,t = 2
, which indicates a decline in the estimation error with the size of the
t + K(1 – ρ )
estimation sample, K, the time horizon, t, and the return correlation, ρ. Consistent with French

21
and Poterba (1991), the relative risk aversion coefficient, γ, is set at 3 for all agents.20

The conservatism coefficient, ξ, for partially-informed agents is set at 0.95, which

reflects their conservative beliefs on the returns of the foreign asset.21 Before observing how

the agents’ estimation errors in μi (i = 1, 2) affect their international portfolio allocation

decisions, we must first investigate the ways in which the size of the estimation sample, K,

affects the variances in their prior beliefs on the mean returns of the US, WI-ex-US and EM

indices. Starting from January 1988, the partially-informed investors formulate their

estimations of the mean returns for the three indices. Given that the data on the US and

WI-ex-US indices are available from January 1970 onwards, the investor has 216 more

observations for these two indices than the EM index.

Figure 1 displays the 95% confidence intervals of the partially-informed agents’

estimates of the means for the US, WI-ex-US and EM index returns. Initially, the confidence

intervals for the EM index are substantially wider than those for the other two indices, which

may well imply that those agents with naïve priors will tend to produce relatively imprecise

estimations of the expected returns. However, the difference in the confidence intervals

between the US and WI-ex-US indices are found to be insignificant, thereby indicating very

similar precision in the information which the US agents have on both the US and WI-ex-US

indices.

<Figure 1 is inserted about here>

We attribute the difference in the patterns of the confidence intervals associated with the

WI-ex-US and EM indices to the longer history of past realizations of returns in the US and

WI-ex-US indices, a factor which also indicates the importance of the advantage that the US

20
As noted by Lundtofte (2006), although the coefficient of relative risk aversion should be greater than 1
(Friend and Blume, 1975), it should not exceed 10 (Mehra and Prescott, 1985); however, some studies attempt to
rationalize observed equity premiums with a higher coefficient of relative risk aversion (such as the coefficient
of 15 used in Brennan and Xia, 2001).
21
By repeating the calibration with different values of ξ, we find that with an increase in the value of ξ , there is a
corresponding increase in the magnitude of the home bias.

22
investors have, which is derived from their cumulative information on the US and WI-ex-US

indices, as compared to the EM index, data on which is only available from January 1988

onwards.

The estimate of mi,t is dynamically formed by partially-informed investors through the

incorporation into the estimation period, K, of recent observations with the passage of time.

We can see that the larger the sample size, the smaller the estimation errors, resulting in the

emergence of the dynamic learning effect with the passage of time. The agents recursively

update their perception of the true expected returns by observing more return realizations,

over time; that is, through the arrival of new information. With more realized returns, the

agents can predict expected returns more precisely, through the reduction in their estimation

errors. Eventually, as the size of the estimation sample, K, becomes substantial, their

estimations of the mean returns will converge towards the true value of the mean.22

Based on Equation (11), the estimation errors relating to the true expected returns of the

home and foreign assets induce hedging demand against the variations in the precision of the

information relative to the expected returns. According to Equation (6), the estimation errors

are dependent upon t, ρ, σi and vi,0 , for i = 1, 2. With the passage of time, the agents acquire

more information on the expected returns, and the partially-informed agents can improve their

estimations of the mean returns through the process of learning. As more observations become

available, there will be corresponding reductions in v1,t and v2,t, with the size of the estimation

sample, K, reflecting the length of the period during which the agents have been active within

the financial markets.

Smaller estimation errors may be associated with a higher degree of correlation between

the home and foreign asset returns, ρ, which may be attributable to more common information

or more common factors relevant to the home and foreign asset returns. The relationship

22
Barry (1974) and Brown (1979) also suggest that the optimal portfolio choice with estimation risk converges
to the choice without estimation risk when the sample size reaches infinity.

23
 2
between vi,t (the estimation error in the mean return) and σ i (the variance in the return) is

dependent upon the magnitude of vi,0 , for i = 1, 2. However, an increase in the return variance

also has a direct effect on the portfolio demand, essentially because it increases the level of

risk arising from investment in the corresponding asset.

The reduced estimation errors relating to the expected returns of the home and foreign

assets also reduces the hedging demand for both of these assets, with these two effects being

found to work in opposite directions for the optimal proportion of the home asset. However,

the conservatism exhibited by investors with regard to foreign investment will tend to hinder,

or depress, their revised estimations of the expected returns on the foreign asset, which will

thereby induce a smaller reduction in the hedging demand for the foreign asset. In general,

with a reduction in the estimation errors, v1,t and v2,t , there will be a corresponding reduction

in the allocation to the home asset.

To illustrate the way in which information quality affects the allocation to the home asset

by conservative partially-informed investors, we further focus on the way in which the home

asset allocation varies with the time horizon t, the point at which the agents determine their

asset allocation decisions. We construct a portfolio based upon the incomplete information

prior to January 1996, and assume that for the trading period from January 1996 onwards, the

fully-informed agents – who bear no estimation errors relating to the mean returns of the

home and foreign asset – will allocate the same amount to home investment, regardless of the

size of their estimation sample, mi,t.

Figures 2a and 2b illustrate the ways in which the allocation to the home asset by

partially-informed agents changes with the time horizon, t, when keeping the sample size

fixed. The partially-informed agents use the data from January 1988 to December 1995 for the

EM index (providing a total of 96 monthly observations) and from January 1970 to December

1995 for the US and WI-ex-US indexes (a total of 312 monthly observations). Based upon the

24
number of observations, K, the correlation between the returns of the home and foreign assets,

ρ, the risk aversion parameter, γ , and the conservatism parameter, ξ, we find that with the

passage of time, the partially-informed agents will steadily reduce their investment in the

home asset, a decline in the home bias which may well be associated with a reduction in the

investors’ estimation risks.

<Figures 2a and 2b are inserted about here>

The dynamic learning scheme enhances the agents’ information quality, thereby reducing

estimation errors with the passage of time. The agents’ portfolio weighting is therefore

increasingly determined by the sum of the first two components in Equation (11). Our results

are consistent with the decline in the home bias observed by both Ahearne et al. (2004) and

Baele et al. (2007), with the estimation errors of the partially-informed agents ultimately

disappearing as the time horizon approaches infinity. This suggests that, over time, the

partially-informed agents’ portfolio strategies will converge to those of the fully-informed

agents, with the home bias relative to foreign investment ultimately disappearing.

Partially-informed US agents exhibit a clear home bias, initially as a means of protecting

themselves against estimation errors in the expected returns on the home asset. However,

given the greater relative volatility of the foreign asset, although such investors should have

greater hedging demand for it, their conservatism in the updating of the estimated mean

returns causes a relatively smaller hedging demand for the foreign asset; thus, the hedging

demand for the home asset, denoted by the third term in Equation (11), ultimately dominates

the hedging demand for the foreign asset, denoted by the fourth term in Equation (11).

The average optimal weights of the home asset for US agents investing in the EM

(WI-ex-US) index are 0.497 (0.4214) for partially-informed agents, and 0.4962 (0.4193) for

fully-informed agents. This result reveals a tendency for conservatism towards the foreign

asset which can clearly help to explain the home bias puzzle. Furthermore, the greater the

25
agents’ risk aversion, then the greater their home bias. In our model, partially-informed

investors do not have perfect knowledge of the return processes, but instead, must estimate

the expected returns of the home and foreign assets using whatever information may be

available to them. As information gradually becomes more readily available, existing beliefs

take into account such new information, ultimately changing to posterior beliefs. By learning

in such a way, the variance in the beliefs of investors will also change, over time, with each

new piece of information.

The fluctuations in the investors’ beliefs and estimation errors relating to the expected

returns are introduced by v1,t and v2,t . An increase in the time horizon, t, leads to a reduction

in the estimation errors relating to μ1 and μ2 , an effect which leads to a reduction amongst

partially-informed investors in the hedging demand for both the home and foreign assets.

Given their conservatism in the updating of the expected returns of the foreign asset, the

reduction in the hedging demand for the foreign asset is dominated by the reduction in the

hedging demand for the home asset; as a result, with an increase in the time horizon, there

will be a corresponding reduction in the proportionate demand for the home asset.

We also find that, across the different markets, different attitudes are exhibited by

partially-informed US agents. In contrast to investment in the emerging markets, we observe a

smaller home bias when the US agents consider investing in a foreign asset in the developed

markets, such as the WI-ex-US index. Since the developed markets have greater market

transparency and a longer history of data on stock returns, the US agents can obtain more

pertinent information. Consistent with Gelos and Wei (2005), we find that partially-informed

investors systematically invest less in those countries whose markets are characterized by

lower levels of transparency.

The numerical results in Figures 1 and 2 suggest that the hedging demands of the agents

are largely dependent upon the conservatism coefficient, ξ, the relative risk aversion

26
parameter, γ, the number of observations, K, the time horizon, t, and the correlation between

the returns of the home and foreign assets, ρ. There is a decline in the overall level of home

bias for conservative agents with K, t and ρ, along with a corresponding increase with the

conservatism coefficient, ξ. A longer estimation period for the formulation of priors or less

stochastic priors enables the partially-informed agents to produce more precise estimates of

the expected returns. The correlation between the returns of the home and foreign asset also

reflects the extent of the co-movement in these returns, which, when they are perfectly

correlated, are found to evolve identically; in such a case, a two-country economy may well be

identical to the economy of a single country.

Our analysis of US agents also applies to foreign, partially-informed agents; even in

equilibrium, partially-informed investors will possess some information on the true expected

returns of the home and foreign assets, whilst conservative agents, as demonstrated by

Equation (7), tend to under-react to new information. Although different types of investors will

infer different expected values for the asset returns, partial equilibrium can be achieved in

Equation (11). Nevertheless, with improvement, over time, in their information set, the

asymmetry in the estimation errors of partially-informed investors relating to the mean returns

of the home and foreign asset will ultimately disappear. The increased information which both

the home and foreign investors are able to obtain on the expected returns of the home and

foreign assets can come either from the learning effects or from the higher correlation

between the returns of the home and foreign assets.

With a lengthening of the estimation period, we observe a smaller confidence interval in

Figure 1, which may well be attributable to the improved quality of the information which

becomes available to investors to form their initial priors. The extent of the under-reaction in

the updating rule on the mean returns of the foreign asset is likely to be gradually reduced as

investors accrue more historical observations upon which to form their priors. When the

27
estimation errors in the expected returns of the home and foreign assets disappear, that is,

when the partially-informed agents gain complete information, then the optimal allocation to

the home asset will be equal to the result obtained in Equation (9). As the agents gain

complete information, the home bias may, therefore, eventually disappear. This gradual

disappearance of the asymmetric expectations of the returns of both the home and foreign

asset is similar to the lack of persistence in heterogeneous beliefs, as noted by Feldman (2007).

Furthermore, heterogeneous beliefs across assets can ultimately converge into homogeneous

beliefs, even with a higher level of correlation between the returns of the foreign and home

assets.

Within our model, the partially-informed agents continue to learn of the true values of

the expected returns, with a general under-reaction to information shocks affecting the returns

of the foreign asset. Figures 2a and 2b shows that the optimal weight applied to the home

asset by partially-informed investors may ultimately converge to that of fully informed agents’

portfolios over time. However, with the arrival of an extreme information shock, investors

will restart the dynamic learning process described in our model; at this time, agents will use

their naïve prior beliefs to re-estimate the true value of the expected returns.

Hence, as predicted in our model, abrupt information shocks may intervene in the

process of investors’ estimations of the mean returns converging to their true values.

Investors’ decisions on the optimal international portfolio cannot reach the benchmark

suggested by the mean-variance strategy, which is why, in the real world, the home bias is still

found to persist amongst investors.

4. CONCLUSIONS
This study investigates the allocation of international assets using an incomplete information

model with consideration of the dynamics of investors’ beliefs and Bayesian uncertainty on

the expected returns of home and foreign market assets. In our model, fully-informed and

28
partially-informed agents each possess different information on the true expected returns. The

fully-informed agents with complete information follow mean-variance optimization to

allocate their wealth globally; in contrast, the estimates of expected returns by

partially-informed agents are subject to estimation errors, essentially because only public

information is available to assist them in their formation of their expected return estimates.

In our model, behavioral bias is incorporated within the decision to undertake

international investment. Thus, in the overall process of updating their estimates of the

unobservable expected returns of the foreign asset, learning bias causes the partially-informed

investors to under-react to recently observed variations within the mean returns process.

Subject to their bounded rationality with regard to foreign investment, such partially-informed

agents will tend to exhibit conservative behavior in the formation of their expectations of the

returns of the foreign asset. As a result, holding more of the home asset ultimately turns out to

be their optimal global portfolio choice.

The quality of information available to investors is improved through the learning effect or

the greater correlation between the home and foreign markets. The importance of the learning

process for the informativeness of securities trading has already been strongly emphasized (Chan

et al., 2007); and indeed, as more information becomes available, partially-informed investors

update their knowledge, learning more about the true value of the expected returns. Furthermore,

the greater correlation between the returns of the home and foreign assets can also relate to

improvements in the quality of information flow across countries, in which case, investors can

estimate the true expected returns of the foreign asset with smaller errors.

Our calibration results reveal that the quality of information also plays an important role,

in terms of helping the agents to form their expected return estimations, and hence, to reach

their decision to hold international assets. As the unfavorable informational position of

partially-informed agents is improved through the learning process, despite their conservatism

29
towards foreign investment, the magnitude of the home bias reduces over time. Our model

provides a potential explanation for the slight reduction in home bias over the past two

decades.23 In a comparison with the single-country models of Brennan (1998) and Lundtofte

(2006), we find that the partial equilibrium in our modified incomplete information economy

with conservative agents performs reasonably well, particularly with regard to explaining the

widely documented lack of international diversification amongst US investors. Unlike Alder

and Dumas (1983), who overlook the changing quality of information, our model explicitly

allows agents to hedge for the changing quality, or precision, of the information relating to the

home and foreign assets.

Two insights thus emerge from the present study. Firstly, if firms or countries can provide

more credible and higher quality information through well-defined corporate governance, they

can attract the attention of international investors, and such investors will exhibit less

conservative behavior in their optimal global investment choices. Secondly, if investors gain

expertise in analyzing and fairly evaluating the financial information gleaned from foreign

countries, they can minimize the effect of incomplete information on their international

portfolio choices.

23
As documented by Ahearne et al. (2004), De Santis and Gerard (2006) and Baele et al. (2007).

30
Appendix: Proofs of Equations (5) and (6)
We solve the filtering problem of the partially informed agent here. Assuming that the system

to be estimated for the partially informed agent is dμi = 0 (i = 1, 2), we can rewrite Equations

(1) and (2) in matrix form:

⎛ dP1,t ⎞
⎜ ⎟ σ
⎟ = ⎛ μ1 ⎞ dt + ⎛⎜ 1 ⎞ ⎛ dZ1 ⎞
0
⎜ P1,t ⎟
⎜ dP2,t ⎜
⎟ ⎝ μ2 ⎠ ⎟ ⎜ σ ρ σ 1 − ρ 2 ⎟ ⎜⎝ dZ 2 ⎟⎠
⎜ ⎟ ⎝ 2 2 ⎠
⎜ P ⎟
⎝ 2,t ⎠

⎛ 1 0 ⎞ ⎛ μ1 ⎞ ⎛ σ1 0 ⎞ ⎛ dZ1 ⎞
=⎜ + ⎜
⎜ σ ρ σ 1 − ρ 2 ⎟⎟ ⎜⎝ dZ 2 ⎟⎠
⎟⎜ μ ⎟ dt . (A.1)
⎝0 1⎠⎝ 2 ⎠ ⎝ 2 2 ⎠

Similarly, we can obtain Equation (4).

According to Liptser and Shiryaev (2001, Theorem 12.7), the optimal estimates m1,t and

m2,t, must take the following matrix form:

⎛ ⎛ dP1,t ⎞ ⎞
−1 ⎜⎜ ⎟ ⎟
⎛ dm1,t ⎞ ⎛ v1,t v12,t ⎞ ⎛ 1 0 ⎞ ⎛ σ 12 σ 1,2 ⎞ ⎜ ⎜ P1,t ⎟ − ⎛ m1,t ⎞ dt ⎟
⎜ ⎟=⎜ ⎟ ⎜ ⎟ ⎟ ⎜ m2,t ⎟ ⎟
⎝ dm2,t ⎠ ⎝ v21,t v2,t ⎠ ⎜⎝ 0 1 ⎟⎠ ⎜⎝ σ 1,2 σ 22 ⎟⎠ ⎜ ⎜ dP
⎜ ⎜ 2,t ⎟ ⎝ ⎠ ⎟
⎜ ⎜ P2,t ⎟ ⎟
⎝⎝ ⎠ ⎠
−1
⎛ v1,t v12,t ⎞ ⎛ σ 12 σ 1,2 ⎞ ⎛ σ 1 0 ⎞ ⎛ dZ1 ⎞
=⎜ ⎟ ⎜⎜ ⎟ ⎜ ⎟ ⎜ ⎟. (A.2)
⎝ v21,t v2,t ⎠ ⎝ σ 1,2 σ 22 ⎟⎠ ⎜⎝ σ 2 ρ σ 2 1 − ρ 2 ⎟⎠ ⎝ dZ 2 ⎠

The matrix of the filtering errors thus evolves according to

⎛ dv1,t dv12,t ⎞
−1
⎜ ⎟ ⎛ v1,t v12,t ⎞ ⎛ 1 0 ⎞ ⎛ σ 12 σ 1,2 ⎞ ⎛ 1 0 ⎞ ⎛ v1,t v12,t ⎞
⎜ dt dt ⎟
= −⎜ ⎟ ⎜ ⎟ ⎜ ⎟, (A.3)
⎜ dv21,t dv2,t ⎟ ⎝ v21,t v2,t ⎠ ⎜⎝ 0 1 ⎟⎠ ⎜⎝ σ 1,2 σ 22 ⎟⎠ ⎜⎝ 0 1 ⎟⎠ ⎝ v21,t v2,t ⎠
⎜ ⎟
⎝ dt dt ⎠
2 I I
where vi,t = E[(μi – mi,t ) | Ft ] for i = 1, 2, and v12,t = v21,t = E[(μ1 – m1,t )(μ2 – m2,t )| Ft]. Using

matrix algebra, we obtain:

31
 ⎛ 1 −ρ ⎞
⎛σ 2
σ 1,2 ⎞
−1 ⎜ σ 2 (1 − ρ 2 ) 2 ⎟
σ 1 σ 2 (1 − ρ )
⎜⎜
1
=⎜ 1 ⎟. (A.4)
2 ⎟⎟
⎝ σ 1,2 σ 2 ⎠ ⎜ −ρ 1 ⎟
⎜ σ σ (1 − ρ 2 ) σ 2 (1 − ρ ) ⎟⎠
2 2
⎝ 1 2

Then, by Equation (A.3), we find

⎛ dv1,t dv12,t ⎞
⎜ ⎟ ⎛ V V12 ⎞
⎜ dt dt ⎟
= − ⎜ 11 ⎟, (A.5)
⎜ dv21,t dv2,t ⎟ ⎝ V21 V22 ⎠
⎜ ⎟
⎝ dt dt ⎠

where

E[( μ1 − m1,t ) 2 Ft I ]2 ρ E[( μ1 − m1,t ) 2 Ft I ]E[( μ1 − m1,t )( μ 2 − m2,t ) Ft I ]

V11 = −
σ 12 (1 − ρ 2 ) σ 1 σ 2 (1 − ρ 2 )

ρ E[( μ1 − m1,t )2 Ft I ]E[( μ1 − m1,t )( μ2 − m2,t ) Ft I ] E[( μ1 − m1,t )( μ2 − m2,t ) Ft I ]2

− + , (A.6)
σ1 σ 2 (1 − ρ 2 ) σ 22 (1 − ρ 2 )

E[( μ1 − m1,t ) 2 Ft I ]E[( μ1 − m1,t )( μ 2 − m2,t ) Ft I ] ρ E[( μ1 − m1,t )( μ 2 − m2,t ) Ft I ]2

V12 = −
σ 12 (1 − ρ 2 ) σ 1 σ 2 (1 − ρ 2 )

ρ E[( μ1 − m1,t ) 2 Ft I ]E[( μ 2 − m2,t ) 2 Ft I ] E[( μ1 − m1,t )( μ 2 − m2,t ) Ft I ]E[( μ 2 − m2,t ) 2 Ft I ]

− + , (A.7)
σ 1 σ 2 (1 − ρ )
2
σ 22 (1 − ρ 2 )
E[( μ1 − m1,t ) 2 Ft I ]E[( μ1 − m1,t )( μ 2 − m2,t ) Ft I ] ρ E[( μ1 − m1,t ) 2 Ft I ]E[( μ 2 − m2,t ) 2 Ft I ]
V21 = −
σ 12 (1 − ρ 2 ) σ 1 σ 2 (1 − ρ 2 )
ρ E[( μ1 − m1,t )( μ 2 − m2,t ) Ft I ]2 E[( μ1 − m1,t )( μ 2 − m2,t ) Ft I ]E[( μ 2 − m2,t ) 2 Ft I ]
− + , (A.8)
σ 1 σ 2 (1 − ρ 2 ) σ 22 (1 − ρ 2 )

and

E[( μ1 − m1,t )( μ 2 − m2,t ) Ft I ]2 ρ E[( μ1 − m1,t )( μ 2 − m2,t ) Ft I ]E[( μ2 − m2,t )2 Ft I ]

V22 = −
σ 12 (1 − ρ 2 ) σ 1 σ 2 (1 − ρ 2 )

ρ E[( μ1 − m1,t )( μ 2 − m2,t ) Ft I ]E[( μ2 − m2,t ) 2 Ft I ] E[( μ 2 − m2,t ) 2 Ft I ]2

− + (A.9)
σ 1 σ 2 (1 − ρ 2 ) σ 22 (1 − ρ 2 )

As noted previously, partially informed agents use processes related to home and foreign asset

prices, as in Equation (A.1), to estimate the expected returns of these assets. However, they

can improve and update their estimates of μi, through the non-zero correlation between home

32
and foreign asset returns (ρ). The priors of μ1 thus link together with the priors of μ2. It is more

difficult to derive comparative statics results, if there is a correlation between priors. To

simplify the analysis, we assume that partially informed agents estimate the home and foreign

asset returns independently, conditional on the correlation between the asset returns.

Therefore, v12,t = v21,t = 0. According to Equation (A.5), the partially informed agents’ filtering

errors in matrix form satisfy the Riccatti equation, as follows:

dv12,t ⎞ ⎛ −v1,t ⎞
2
⎛ dv1,t 0
⎜ ⎟ ⎜ 2 ⎟
dt ⎟ ⎜ σ 1 (1 − ρ )
2
⎜ dt = ⎟. (A.10)
⎜ dv21,t dv2,t ⎟ ⎜ −v2,t ⎟
2

⎜ ⎟ ⎜ 0 ⎟
⎝ dt dt ⎠ ⎜⎝ σ 22 (1 − ρ 2 ) ⎟⎠

By solving Equation (A.10), we obtain Equation (6):

⎛ v1,0σ 12 (1 − ρ 2 ) ⎞
⎜ 0 ⎟
0 ⎞ ⎜ tv1,0 + σ 1 (1 − ρ )
2 2
⎛ v1,t ⎟
⎜ ⎟ = .
⎝ 0 v2,t ⎠ ⎜ v2,0σ 2 (1 − ρ ) ⎟⎟
2 2
⎜ 0
⎜ tv2,0 + σ 22 (1 − ρ 2 ) ⎟⎠
⎝

Then, we can derive Equation (5) as follows:

⎛ 1 −ρ ⎞
⎛ dm1,t ⎞ ⎛ v1,t ⎜
0 ⎞ σ 1 (1 − ρ 2 )
2
σ 1σ 2 (1 − ρ 2 ) ⎟ ⎛ σ 1 0 ⎞ ⎛ dZ1 ⎞
⎜ ⎟=⎜ ⎟⎜ ⎟⎜ ⎟⎜ ⎟
⎝ dm2,t ⎠ ⎝ 0 v2,t ⎠ ⎜ −ρ 1 ⎟ ⎜⎝ σ 2 ρ σ 2 1 − ρ 2 ⎟⎠ ⎝ dZ 2 ⎠
⎜ σ σ (1 − ρ 2 ) σ 22 (1 − ρ 2 ) ⎟⎠
⎝ 1 2
⎛ v1,t − ρ v1,t⎞
⎜ ⎟
⎜ σ1 σ 1 1 − ρ ⎟ ⎛ dZ1 ⎞
2

=⎜ ⎟ ⎜ dZ ⎟ .
v2,t
⎜ 0 ⎟⎝ 2 ⎠
⎜ 2 ⎟
σ 2 1− ρ ⎠
⎝

33
References
Ahearne, A., Griever, W., Warnock, F., 2004, Information costs and home bias: An analysis of
US holdings of foreign equities, Journal of International Economics, 62, 313-336.
Alder, M., Dumas, B., 1983, International portfolio choice and corporation finance: A
synthesis, Journal of Finance, 38, 925-984.
Back, K., 2004, Incomplete and asymmetric information in asset pricing theory, in M. Frittelli
and W. Runggaldier, eds., Stochastic Methods in Finance, Springer.
Baele, L., Inghelbrecht, K., 2006, Structural versus temporary drivers of country and industry
risk, Working Paper.
Baele, L., Pungulescu, C., Ter Horst, J., 2007, Model uncertainty, financial market integration
and the home bias puzzle, Journal of International Money and Finance, 26, 606-630.
Barberis, N., 2000, Investing for the long run when returns are predictable, Journal of Finance,
55, 307-349.
Barberis, N., Huang, M., Santos, T., 2001, Prospect theory and asset prices, Quarterly Journal
of Economics, 116, 1-54.
Barry, C., 1974, Portfolio analysis under uncertain means, variances, and covariances, Journal
of Finance, 29, 515-522.
Berrada, T., 2009, Bounded rationality and asset pricing with intermediate consumption,
Review of Finance, 13, 693-725.
Brennan, M., 1998, The role of learning in dynamic portfolio decisions, European Finance
Review, 1, 295-306.
Brennan, M., Cao, H., 1997, International portfolio investment flows, Journal of Finance, 52,
1858-1880.
Brennan, M., Xia, Y., 2001, Stock return volatility and equity premium, Journal of Monetary
Economics, 47, 249-283.
Brown, S., 1979, The effect of estimation risk on capital market equilibrium, Journal of
Financial and Quantitative Analysis, 14, 215-220.
Burger, J., Warnock, F., 2007, Foreign participation in local currency bond markets, Review
of Financial Economics, 16, 291-304.

Chan, K. Menkveld, A., Yang, Z., 2007, The informativeness of domestic and foreign
investors’ stock trades: Evidence from the perfectly segmented Chinese market, Journal of
Financial Markets, 10, 391-415.

34
Daniel, K., Hirshleifer, D., Subrahmanyam, A., 2001, Overconfidence, arbitrage, and
equilibrium asset pricing, Journal of Finance, 56, 921-965.

De Santis, R.A., Gerard, B., 2006, Financial integration, international portfolio choice and the
European Monetary Union, ECB Working Paper 626.
Detemple, J., 1986, Asset pricing in a production economy with incomplete information,
Journal of Finance, 41, 383-391.
Detemple, J., 1991, Further results on asset pricing with incomplete information, Journal of
Economic Dynamics and Control, 15, 425–453.
Dothan, M.U., Feldman, D., 1986, Equilibrium interest rates and multiperiod bonds in a
partially observable economy, Journal of Finance, 41, 369-382.
Feldman, D., 1989, The term structure of interest rates in a partially observable economy,
Journal of Finance, 44, 789–812.
Feldman, D., 2003, The term structure of interest rates: Bounded or falling? European Finance
Review (Review of Finance), 7, 103-113.
Feldman, D., 2007, Incomplete information equilibria: Separation theorems and other myths,
Annals of Operations Research, 151, 119-149.
French, K., Poterba, J., 1991, Investor diversification and international equity markets,
American Economic Review, 81, 222-226.
Friend, I. and Blume, M., 1975, The demand for risky assets, American Economic Review, 65,
900-922.
Gehrig, T., 1993, An information-based explanation of the domestic bias in international
equity, Scandinavian Journal of Economics, 95, 97-109.
Gelos, R. G., Wei, S., 2005, Transparency and international portfolio holdings, Journal of
Finance, 60, 2987-3020.
Griffin, J., Karolyi, A., 1998, Another look at the role of the industrial structure of markets for
international diversification strategies, Journal of Financial Economics, 50, 351-373.
Grinblatt, M., Keloharju, M., 2001, How distance, language, and culture influence
stockholdings and trades, Journal of Finance, 56, 1053-1073.
Hasan, I., Simaan, Y., 2000, A rational explanation for home country bias, Journal of
International Money and Finance, 19, 331-361.
Heston, S., Rouwenhorst, K., 1994, Does industrial structure explain the benefits of
international diversification? Journal of Financial Economics, 36, 3-27.
Huberman, G., 2001, Familiarity breeds investment, Review of Financial Studies, 14,

35
 659-680.

Kang, J.-K., Stulz, R., 1997, Why is there a home bias? An analysis of foreign portfolio equity
ownership in Japan, Journal of Financial Economics, 46, 3-28.

Karlsson, A., Norden, L., 2007, Home sweet home: Home bias and international
diversification among individual investors, Journal of Banking and Finance, 31, 317-333.
Karolyi, A., Stulz, R., 2003, Are financial assets priced locally or globally? in: G.M.
Constantinides, M. Harris, and R. M. Stulz, eds., Handbook of the Economics of Finance,
Chapter 16, 975-1020.
Kilka, M., Weber, M., 2000, Home bias in international stock return expectations, Journal of
Behavioral Finance, 1, 176-192.
Lewis, K., 1999, Trying to explain home bias in equity and consumption, Journal of
Economic Literature, 37, 571-608.

Liptser, R., Shiryaev, A., 2001, Statistics of Random Processes, I and II, Berlin:
Springer-Verlag.
Lundtofte, F., 2006, The effect of information quality on optimal portfolio choice, Financial
Review, 41, 157-185.
Lundtofte, F., 2008, Expected life-time utility and hedging demands in a partially observable
economy, European Economic Review, 52, 1072-1096.
Lundtofte, F., 2009, Can an ‘estimation factor’ help explain cross-sectional returns? Journal of
Business, Finance and Accounting, 36, 705-724.
Lutje, T., Menkhoff, L., 2007, What drives home bias? Evidence from fund managers’ views,
International Journal of Finance and Economics, 12, 21-35.
Massa, M., Simonov, A., 2005, Is learning a dimension of risk, Journal of Banking and
Finance, 29, 2605-2632.
Massa, M., Simonov, A., 2006, Hedging, familiarity and portfolio choice, Review of Financial
Studies, 19, 633-685.
Mehra, R., Prescott, E., 1985, The equity premium: A puzzle, Journal of Monetary Economics,
15, 145-161.
Merton, R., 1980, On estimating the expected return on the market: An explanatory
investigation, Journal of Financial Economics, 8, 323-361.
Morse A, Shive S., 2006, Patriotism in your portfolio: Country loyalty and the home bias,
Working Paper.

36
Portes, R., Rey, H., 2005, The determinants of cross-border equity flows, Journal of
International Economics, 65, 269-296.
Riedel, F., 2000, Decreasing yield curves in a model with an unknown constant growth rate,
European Finance Review (Review of Finance), 4, 51-67.
Sercu, P., Vanpee, R., 2007, Home bias in international equity portfolios: A review, Working
Paper.
Shiller, R., Kon-Ya, F., Tsutsui Y., 1996, Why did the Nikkei crash? Expanding the scope of
expectations data collection, Review of Economics and Statistics, 78, 156-164.
Ueda, M., 1999, Incomplete observation, filtering, and the home bias puzzle, Economics
Letters, 62, 75-80.
Veronesi, P., 2000, How does information quality affect stock returns? Journal of Finance, 55,
807-837.
Zhou, C., 1998, Dynamic portfolio choice and asset pricing with differential information,
Journal of Economic Dynamics and Control, 22, 1027-1051.
Ziegler, A., 2000, Optimal portfolio choice under heterogeneous beliefs, Review of Finance, 4,
1-19.

37
Table 1. International Home Bias in Equity Portfolios, December 2005
Portfolio investment data come from the IMF’s CPIS. The market capitalization data are from
the World Federation of Exchange. The equity home bias can be calculated by deducting the
ratio of total domestic market capitalization from the proportion of domestic equity in a
country’s portfolio.

Percentage of Equity
Market Capitalization,
Country Portfolio in Equity Home Bias
Percentage of Total
Domestic Equities
Argentina 0.11 82.64 82.53
Australia 1.83 83.57 81.74
Austria 0.29 57.79 57.51
Belgium 0.65 49.69 49.04
Brazil 1.08 99.26 98.18
Canada 3.37 75.59 72.22
Chile 0.31 84.97 84.66
Colombia 0.11 98.00 97.88
Czech Republic 0.12 88.75 88.62
Denmark 0.42 61.91 61.49
Egypt 0.18 98.81 98.63
Finland 0.54 65.87 65.32
France 3.99 68.53 64.53
Germany 2.77 57.41 54.64
Greece 0.33 93.35 93.02
Hong Kong 2.40 80.24 77.85
Hungary 0.07 90.59 90.51
India 2.43 100.00 97.57
Indonesia 0.18 99.85 99.67
Israel 0.28 91.34 91.06
Italy 1.81 57.22 55.41
Japan 17.13 94.16 77.04
Korea 1.63 97.39 95.76
Malaysia 0.41 99.02 98.61
Mexico 0.54 98.14 97.60
Netherlands 1.35 32.57 31.23
New Zealand 0.09 59.09 59.00
Norway 0.43 51.94 51.51
Philippines 0.09 99.65 99.56
Poland 0.21 97.90 97.69
Portugal 0.15 72.11 71.96
Russia 1.35 99.94 98.58
Singapore 0.58 70.68 70.09
South Africa 1.25 88.90 87.66
Spain 2.18 86.21 84.03
Sweden 1.00 60.82 59.82
Switzerland 2.12 59.67 57.54
Thailand 0.28 98.95 98.67
Turkey 0.37 99.93 99.57
United Kingdom 6.94 63.12 56.17
United States 38.60 82.17 43.57
Venezuela 0.02 94.29 94.27

38
Table 2. Summary Statistics for Equity Returns
The data cover monthly indexes in U.S. dollars, including the MSCI United States (US) Index,
World Index ex USA (WI-ex-US) Index, and Emerging Markets (EM) Index. The sample
period spans January 1970 to December 2006 for the US and WI-ex-US Indexes and January
1988 to December 2006 for the EM Index.

Panel A: Descriptive Statistics of Returns

EM WI-ex-US US
Mean 0.0097 0.0068 0.0058
Std. Dev. 0.0669 0.0478 0.0439
Kurtosis 3.2223 1.1231 2.6280
Skewness - 1.0355 - 0.4310 - 0.5862
Sharpe Ratio 0.1449 0.1423 0.1329
Observations 228 444 444

Panel B: Correlation Coefficient

EM WI-ex-US US
EM 1
WI–ex-US 0.6083* 1
*
US 0.5973 0.5775 1
*Sample correlation coefficients are calculated for the period from January 1988 to December
2006.

39
Figure 1. The 95% Confidence Intervals of the Partially Informed Agent’s Estimates of the True Value of Means
(a) The US Index (b) The WI-ex-US Index
0.014 0.014

0.012 0.012
0.010
0.010
0.008
0.008
0.006
0.006
0.004
0.004
0.002

0.000 0.002
1 101 201 301 401 501 601 701 801 901 1001
(0.002) 0.000
1 101 201 301 401 501 601 701 801 901 1001
(0.004)
Estimation Period in Months Estimation Period in Months

(c) The EM Index

0.150

0.100

0.050

0.000
1 101 201 301 401 501 601 701 801 901 1001
(0.050)

(0.100)

(0.150)
Estimation Period in Months

Note: One thousand and one post-initial observations. Initially, the estimation sample for the US and WI-ex-US indexes is January 1970 to January 1988, while the
estimation sample is January 1988 for the EM index. As time goes by, the partially informed investor updates her interval estimate for the expected return as more
observations become available. So the investor updates the estimation sample and changes the interval estimate every month. The true value of expected returns is 0.0058
for the US index, 0.0068 for the WI-ex-US index, and 0.0097 for the EM index. The (monthly) standard deviation of returns is set to 0.0439 for the US index, 0.0478 for
the WI-ex-US index, and 0.0669 for the EM index.

40
Figure 2. (a), (b) Agents' Optimal Portfolio Weights
(a) Weight of the US index against the EM index
0.4975

0.4970 P‐3

0.4965
θ

0.4960 F‐3

0.4955
0 100 200 300 400 500 600 700 800 900 1000

Time Horizon in Months

(b) Weight of the US index against the WI-ex-US index

0.422

0.421 P‐3

0.420
θ

0.419 F‐3

0.418
0 100 200 300 400 500 600 700 800 900 1000

Time Horizon in Months

Note: One thousand post-initial observations, where the initial date is January 1996. The fully
informed U.S. agent invests in the EM index, her optimal proportion to the home asset is constant at
0.4961 (red solid line). On the other hand, the fully informed U.S. agent invests in the WI-ex-US
index, her optimal portfolio choice to the home asset keeps constant at 0.4193 (red solid line). The
parameter values used for the determination of the partially informed agent’s optimal portfolio weight
(blue dotted line) are μ1 = 0.0058, σ1 = 0.0439 for the US index, μ2 = 0.0068, σ2 = 0.0478 for the
WI-ex-US index, μ2 = 0.0097, σ2 = 0.0669 for the EM index, ρ = 0.5973 for the portfolio of US and
EM indexes, and ρ = 0.5775 for the portfolio of US and WI-ex-US indexes.

41