Vous êtes sur la page 1sur 19

Received: 30 June 2017 | Revised: 28 May 2018 | Accepted: 1 June 2018

DOI: 10.1002/fut.21941

RESEARCH ARTICLE

Correlation risk and international portfolio choice

Nicole Branger1 | Matthias Muck2 | Stefan Weisheit2

1
Finance Center Muenster, University
of Muenster, Universitätsstr.
Variance‐covariance risk of the exchange rate is highly relevant for interna-
14‐16, Muenster, Germany tional investors. This paper addresses optimal asset allocation with stochastic
2
Chair of Banking and Financial Control, variances and covariances in a Wishart Affine Stochastic Correlation (WASC)
University of Bamberg, Bamberg,
model in incomplete and complete markets. We show that the (hedging)
Germany
demand for exchange rate variance‐covariance risk can differ significantly
Correspondence between international investors. Local correlations with the exchange rate can
Matthias Muck, Chair of Banking and
affect the utilities of international investors differently while the impact of
Financial Control, University of Bamberg,
Kärntenstr. 7, 96045 Bamberg, Germany. correlations between stocks can be symmetric. Depending on the current local
Email: matthias.muck@uni-bamberg.de exchange rate correlations domestic investors can benefit more or less than
foreign investors from international trading.

KEYWORDS
International asset allocation, stochastic correlation, Wishart processes, dynamic trading strategies,
derivatives

JEL CLASSIFICATION
G11, G13

1 | INTRODUCTION

Empirical evidence suggests that return variances and return correlations are stochastic, and that stochastic second
moments are priced. This has been documented within the stock market, across asset classes, and also on international
markets.1 Stochastic second moments have far‐reaching implications for the asset allocation of investors because
expected returns and risk exposures are no longer deterministic, and investment opportunity sets become stochastic.
While stochastic variances have attracted a lot of attention in the literature, approaches to the modeling of stochastic
correlations are relatively new. Within the class of affine models, so‐called Wishart processes allow to specify the joint
dynamics of variances and covariances directly.2 Examples in the literature on portfolio choice are Buraschi, Porchia
and Trojani (2010) and Da Fonseca, Grasselli and Ielpo (2011) who consider the analytically tractable Wishart Affine
Stochastic Correlation (WASC) model.
In this paper we study the asset allocation of international investors. We are interested in how stochastic covariances
and stochastic correlations affect the portfolio choice of domestic and foreign investors. We use a Wishart process to
model the variances and covariances of domestic and foreign stock markets and the foreign exchange rate, and compare
the optimal asset allocation decisions of domestic and foreign investors. Furthermore, we consider benefits from
international trading and compare utility improvements of domestic and foreign investors in particular for different
levels of correlation.

1
Bollerslev, Engle and Wooldridge (1988), Longin & Solnik (1995), Ball & Torous (2000) and Goetzmann, Li and Rouwenhorst (2005) document that international market correlations vary dynamically
over time. Moreover, Driessen, Maenhout and Vilkov (2009) find that correlation risk is priced based on index and individual option prices. The results provided by Krishnan, Petkova and Ritchken
(2009) and Driessen, Maenhout and Vilkov (2012) imply a sufficiently large, negative risk premium for correlation risk. Further studies that deal with the pricing of variance and correlation risk are
Hollstein & Simen (2017), Buss, Schoenleber and Vilkov (2018).
2
Wishart processes were first studied by Bru (1991). Gourieroux (2006) and Da Fonseca, Grasselli and Tebaldi (2007) study the pricing of derivative claims in Wishart frameworks.

128 | © 2018 Wiley Periodicals, Inc. wileyonlinelibrary.com/journal/fut J Futures Markets. 2019;39:128–146.


BRANGER ET AL. | 129

The analysis is done both in an incomplete market where the investors can only trade stocks and bonds, and in a
complete market. In a complete market arbitrary exposures to variance‐covariance risk can be attained by trading
derivatives. As it is, for example, discussed by Liu & Pan (2003) this can lead to significant utility improvements for
investors. Mathematically, the particular types of contracts that complete the market do not matter as long as the
derivatives provide an exposure to variance‐covariance risk. In practice, though, it might be advantageous to trade
derivatives whose prices are directly connected to elements of the variance‐covariance matrix. Natural candidates to
trade stock and exchange rate variance risk are stock options and exchange rate options. Quanto contracts and options
on several assets provide an exposure to covariance risk. Besides options, one might also trade futures and swaps on
variances, covariances, and correlations. An advantage of the WASC model is that its affine structure results in quasi
closed form solutions for standard derivatives which allow for an efficient calculation of prices and Greeks. In practice,
this facilitates the application of the model discussed in this paper.
Our paper makes several contributions. First, we consider the joint dynamics of foreign and domestic stock prices
in their local currencies and the (direct) exchange rate when variances, covariances, and correlations are stochastic.
Given this starting point, we derive the investment opportunity sets of domestic and foreign investors who can trade in
international stocks and bonds. The stochastic variance‐covariance matrix is driven by a Wishart process.3
Second, we determine the optimal portfolios of international investors and their utility gains from trading
international stocks and bonds. We assume that investors have constant relative risk aversion (CRRA). While the
demand for stocks is similar for domestic and foreign investors, the demand for foreign bonds – which allow to trade
exchange rate risk – differs significantly between the investors. We also find that the benefits from international trading
are different for domestic and foreign investors. Utility gains depend on the current levels of variances and correlations.
In our example, both certainty equivalent returns increase in the correlation between the stocks, but decrease in the
correlations between the stocks and the exchange rate. Furthermore, the difference between the certainty equivalent
returns can change sign depending on the latter correlations.
Third, we assess the impact of market completion, that is, we consider a market in which international investors have
access to derivatives. Investors can disentangle stock and bond market risk from variance‐covariance risk and exploit risk
premia for stochastic variances and covariances. In our numerical example we find that, in contrast to, for example, the
exposure to the covariance between the stocks, the exposures to the covariance between stocks and the exchange rate are
significantly different and even change sign. The benefits from trading depend on the stochastic correlations. As in the
incomplete market, changes in stock correlations have a symmetric impact on domestic and foreign investors, while changes
in the correlation between stocks and exchange rates might have different consequences.
Our paper is related to several strands of the literature. Following Merton (1969, 1971, 1973) numerous papers deal
with asset allocation problems that take additional risk factors like stochastic volatility or jumps into account. Examples
include Liu, Longstaff and Pan (2003), Chacko & Viceira (2005) and Liu (2007). The implications of market
completeness are explored by, for example, Liu & Pan (2003), Branger, Schlag and Schneider (2008), Muck (2010),
Egloff, Leippold and Wu (2010) and Da Fonseca et al. (2011). Furthermore, Branger, Muck, Seifried and Weisheit (2017)
consider the impact of jumps in variances and covariances in a WASC model for stocks.
Alternatives to the WASC model exist in the literature as well. For example, Engle (2002) proposes the Dynamic
Conditional Correlation (DCC) model. This approach facilitates the econometric estimation of the dynamics of the
correlation matrix. However, in this framework we cannot distinguish between complete and incomplete markets.
Moreover, the pricing of derivatives is more involved. Another possibility is to consider latent state variables (different
from a Wishart process) to describe the variances and covariances. This is, for example, the case in the multi‐Heston
model of De Col, Gnoatto and Grasselli (2013). The multi‐Heston models also results in quasi closed form solutions for
derivative prices. However, covariances are usually not represented by distinct state variables in this approach, but are
driven by the same state variables that determine the variances.
Our paper is also related to the literature on international asset allocation and the benefits from international
investing. Empirical evidence on gains from international equity diversification is provided by, among others, Kaplanis
& Schaefer (1991), Bekaert & Urias (1996) and Li, Sarkar and Wang (2003). In the spirit of Adler & Dumas (1983), a
number of papers study the implications of international asset allocation in dynamic models.4 Lioui & Poncet (2003)

3
Our setup nests the special case in which stock prices and exchange rates are driven by a WASC model.
4
Adler & Dumas (1983) show that international portfolio selection is closely related to portfolio choice problems under inflation risk. The latter usually consider real prices obtained from dividing
nominal prices by a price index, which is identical to dividing prices given in a measurement currency by the exchange rate. Examples for models with inflation risk are Brennan & Xia (2002) and
Munk, Sørensen and Nygaard Vinther (2004).
130 | BRANGER ET AL.

analyze international asset allocation with interest rate risk. Smedts (2004) explores the impact of eliminating the
exchange rate on international portfolios. Similar to our framework, she assumes that investors can trade domestic and
foreign stocks and bonds. Larsen (2010) analyzes a dynamic asset allocation model and shows that investors suffer a
loss from not investing internationally. In contrast to these studies, our focus is on stochastic variances and correlations
in an international market. Ang & Bekaert (2002) study the impact of changing correlations caused by regime switches
in an international asset allocation model. Similarly, Escobar, Ferrando and Rubtsov (2015) study an international
Principal Component Stochastic Volatility (PCSV) model. Our model differs from these approaches as we assume that
the joint dynamics of return variances and correlations are driven by a diffusion Wishart process.
The remainder of the paper is organized as follows: Section 2 introduces the model. In Section 3 we discuss
the asset allocation problem for domestic and foreign investors when markets are incomplete. In Section 4 we calibrate
the model and analyze optimal portfolios as well as benefits from trading in a numerical example. Section 5 deals with
the investment problem when markets are complete. Section 6 concludes.

2 | THE M O DE L

We consider an arbitrage‐free economy in continuous‐time with two countries, the domestic and the foreign one. The
exchange rate is quoted in direct terms, that is, it denotes the amount of the domestic currency that must be paid for one
unit of the foreign currency. For convenience and to simplify the exposition, we call the domestic country USA (with
currency USD) and the foreign country Europe (with currency EUR).
In each country, there is a locally risk‐free money market account and a risky stock. The investors have access to
domestic and foreign financial markets and can thus trade all four assets. A similar setup has been studied by Adler &
Dumas (1983). In line with empirical evidence, we extend their approach by introducing stochastic variances and
stochastic covariances. We follow Buraschi et al. (2010) and Da Fonseca et al. (2011) and rely on the WASC model. This
model considers a state matrix which can be interpreted as variance‐covariance matrix of the asset returns. This allows
to model variance‐covariance risk directly.
In the following we discuss the model for asset returns. We proceed as follows. In Section 2.1 we introduce the model
for international equity and exchange rate returns. Section 2.2 derives the investment opportunity sets faced by a
domestic (US) and a foreign (EU) investor. Section 2.3 defines international risk premia.

2.1 | Return dynamics


International investors may trade domestic and foreign stocks and money market accounts. The money market
accounts earn a constant riskless interest rate r US and r EU in the US and in Europe, respectively. The European (US)
money market account is risky from a US (European) perspective due to exchange rate risk. In the following F denotes
the USD/EUR exchange rate, that is, the amount of USD that one has to pay for one EUR. Moreover, in each country, a
stock (which represents the whole stock market) is traded. SUS and S EU denote the USD prices of the US and the
SUS S EU
European stock, respectively. The corresponding EUR prices are SUS * = F and S EU * = F . We assume that there are
no dividend payments.
We start by modelling the prices of the US stock (in USD), the European stock (in EUR), and the foreign exchange
rate. The prices are collected in the vector V Market = (SUS , S EU *, F )′. The price dynamics under the physical measure 
are given by

dVtMarket = diag (VtMarket )(μtMarket dt + α Xt dZt ), (1)

where diag(VtMarket ) is a diagonal matrix with the elements of the price vector V Market on the main diagonal. Z ∈ 3 is a
vector Brownian motion, μtMarket denotes the expected returns, and X ∈ 3 × 3 drives the variance‐covariance matrix of
returns. Moreover, the matrix α ∈ 3 × 3 is constant.
The matrix X follows the Wishart‐process

dXt = (ΩΩ′ + MXt + Xt M ′)dt + Xt dBt Q + Q′dB′t Xt , (2)


BRANGER ET AL. | 131

where B ∈ 3 × 3 is a matrix‐valued Brownian motion. Ω, M , Q ∈ 3 × 3 are square matrices, and we assume that Ω is
invertible. The Wishart process describes the dynamics of positive semi‐definite matrices. It was first analyzed by Bru
(1991), and represents the matrix‐analogue of the Cox, Ingersoll and Ross (1985) process. Its long‐run average X∞
follows from

ΩΩ′ + MX∞ + X∞ M ′ = 0. (3)

With this definition, the dynamics of X can be written as

dXt = [−M (X∞ − Xt ) − (X∞ − Xt ) M ′]dt + Xt dBt Q + Q′dB′t Xt . (4)

Throughout we require that ΩΩ′ = kQ′Q for k > 2, which guarantees that X is positive semi‐definite.5
The variance‐covariance matrix of stock and exchange rate returns is given by Y Market = αXα′.6 The elements of the
correlation matrix ρ̄ ∈ 3 × 3 of returns are

YtMarket
, ij
ρt̄ , ij ≡ , (5)
YtMarket
, ii YtMarket
, jj

where Yij and ρ̄ij are the elements {i , j} of Y and ρ̄, respectively. The Wishart process thus allows not only for stochastic
variances and covariances, but also for stochastic correlations.
Our model can also capture non‐zero correlations between returns and return variances which determine the shapes
of the volatility smiles. The Brownian motions Z and B which drive returns and their variance‐covariance matrix,
respectively, are related via

dZt = dBt ρ + dWt 1 − ρ′ρ (6)

with ρ ∈ 3 such that ρ′ρ ≤ 1. W ∈ 3 is a vector Brownian motion independent of B . The correlation between the
return dViMarket ∕ViMarket and its variance Yii (leverage) is given by

dVtMarket (αQ′ρ)i
Corr ⎡ ⎤
,i
⎢ V Market , dYt , ii⎥ = (αQ′Qα′)ii
. (7)
⎣ t,i ⎦

In contrast to the return correlations, which are stochastic, the correlations between asset returns and their local
variances are constant.

2.2 | Investment opportunity sets


In this subsection we characterize the available investment opportunity sets. We begin with the US investor who can
trade three risky securities including the US stock, the European stock, and the European risk-free money market
account which earns the European risk-free rate of interest (EU bond in the following). The latter is risky from the US
perspective because it is exposed to exchange rate risk, and its USD price is given by P EU . Furthermore, the US investor
can invest in a risk-free US money market account which earns the US risk-free rate of interest (US bond in the
following). The vector V US = (SUS , S EU , P EU ) ′ collects the prices of the risky securities denoted in USD. Its dynamics
are provided by the following Lemma.

Lemma 1 From the perspective of the US investor, the dynamics of the risky securities denoted in USD are given by

5
The rather strong condition on Ω is most commonly used in the relevant portfolio choice literature (see, e.g., Leippold & Trojani (2010) or Buraschi et al. (2010)). Cuchiero, Filipović, Mayerhofer and
Teichmann (2011) show that it can be relaxed to ΩΩ′ − 2Q′Q ∈ S3+ , where S3+ is the cone of symmetric, positive semi‐definite matrices in 3 × 3 .
6
In the special case when α is the identity matrix, Equations 1 and 2 characterize the Wishart Affine Stochastic Correlation (WASC) model in which the variance‐covariance matrix of returns is Y = X .
132 | BRANGER ET AL.

dVtUS = diag (VtUS )(μtUS dt + βUS Xt dZt ), (8)

where

⎛1 0 0⎞
βUS = ⎜ 0 1 1 ⎟ α.
⎝0 0 1⎠

The variance‐covariance matrix Y US of risky securities is

Y US = βUSXβUS′.

EU
Proof The USD prices of the European stock and bond are given by S EU = S EU *F and PtEU = e r tFt , respectively.
Equation (1) and an application of Itô’s Lemma yields Equation (8). The variance‐covariance matrix follows
immediately. □

The expected return μUS will be given in Section 2.3 where we discuss the market prices of risk. As above, the return
correlations are stochastic, while the correlations between returns and their variances are constant.
Lemma 1 shows that our model nests a WASC model for the USD denominated returns. In this case the matrix βUS is
equal to the identity matrix, the matrix α is

−1
⎛1 0 0⎞ ⎛1 0 0⎞
α = ⎜ 0 1 1 ⎟ = ⎜ 0 1 − 1⎟ ,
⎝0 0 1⎠ ⎝ 0 0 1⎠

and the variance‐covariance matrix of USD denominated returns reduces to Y US = X .


The investment opportunity set of a European investor can be described in a similar way. From his perspective the
prices of the stocks and the US bond are risky. Let PUS * be the price of the US bond in EUR and let
V EU = (SUS *, S EU *, PUS *)′ collect the EUR prices of the risky securities. Then the following Lemma provides the EUR
dynamics of these risky securities.

Lemma 2 From the perspective of the EU investor, the dynamics of the risky securities denoted in EUR are given by

dVtEU = diag (VtEU )(μtEU dt + β EU Xt dZt ), (9)

where

⎛ 1 0 − 1⎞ ⎛ 1 0 − 1⎞
β EU = ⎜ 0 1 0 ⎟ α = ⎜ 0 1 − 1⎟ βUS .
⎝ 0 0 − 1⎠ ⎝ 0 0 − 1⎠

The variance‐covariance matrix Y EU of the risky returns is

Y EU = β EU Xβ EU ′.

S US e rUSt
Proof The EUR prices of the US stock and bond are given by SUS * = F and PtUS * = F . Equation (1) and an
t
application of Itô’s Lemma yields Equation (9). The variance‐covariance matrix follows immediately. □
BRANGER ET AL. | 133

2.3 | Pricing kernels and risk premiums


We now turn to the market prices of risk and thus to the premia paid for stock price risk, variance‐covariance risk, and
correlation risk. M US is the pricing kernel from the perspective of the US investor. Its dynamics are exogenously
given by7

dMtUS
= −r US dt − tr (ηUS ′ Xt dBt ) − ξ US′ Xt dWt . (10)
MtUS

where the constant coefficients ηUS ∈ 3 × 3 and ξ US ∈ 3 describe the exogenously specified market prices of risk for B
and W , respectively.8
The risk premiums of the assets follow from the covariances between the asset returns and the pricing kernel. From
the perspective of the US investor, the risk premiums for the stocks and the EU bond (in USD) are

μtUS − r US 1 = βUSXt ΛUS , (11)

where

ΛUS = ηUS ρ + ξ US 1 − ρ′ρ . (12)

This gives the expected return μtUS of the USD denominated returns in Equation (8).
Under the equivalent risk neutral measure US the state matrix X again follows a Wishart process

Xt dBt Q + Q′dBt
US US′
dXt = (ΩΩ′ + ℳUSXt + Xt ℳUS′)dt + Xt , (13)

where ℳUS = M − Q′ηUS′. The instantaneous risk premium for variance‐covariance risk results from the dynamics of
the state matrix X under  and US in Equations 2 and 13 and it is given by

1
(Et [dYtUS] − Et [dYtUS]) = βUS (Xt ηUS Q + Q′ηUS′Xt ) βUS′.
US
CovRPtUS ≡ (14)
dt

Next, we address the pricing kernel from a European perspective. The corresponding European pricing kernel MtEU
is given by

MtUS Ft
MtEU = . (15)
F0

By a standard argument the drift of the exchange rate in Equation (1) is given by

dM US dF ⎤
μ3Market = r US − r EU − COV ⎡ US , .

⎣ M F ⎥

Applying Ito’s Lemma to Equation (15) then yields

dMtEU
= −r EU dt − tr (η EU ′ Xt dBt ) − ξ EU ′ Xt dWt , (16)
MtEU

where

7
See, e.g., Da Fonseca et al. (2007).
8
The specification of the market prices of risk guarantees analytical tractability for the solution to the portfolio allocation problem. Generalized market prices of risk in Wishart‐based models are
suggested by Chiarella, Hsiao and To (2011) or Gruber, Tebaldi and Trojani (2014).
134 | BRANGER ET AL.

η EU = ηUS − βUS′δρ′ ξ EU = ξ US − 1 − ρ′ρ βUS′δ , (17)

and δ = (0 0 1) ′. From the perspective of the EU investor, the risk premia on the US stock, the EU stock, and the US
bond (in EUR) are thus

μtEU − r EU 1 = β EU Xt ΛEU , (18)

where

ΛEU = η EU ρ + ξ EU 1 − ρ′ρ = ΛUS − βUS′δ . (19)

The instantaneous risk premium for variance‐covariance risk Y EU is

1
(Et [dYtEU ] − Et [dYtEU ]) = β EU (Xt η EU Q + Q′η EU ′Xt ) β EU ′
EU
CovRPtEU ≡ (20)
dt

Finally and for the sake of completeness, we note that the expected returns in Equation (1) are affine in the state
matrix X and given by

US US US
⎛ r + (β Xt Λ )1 ⎞
μtMarket = ⎜ r EU + (β EU Xt ΛEU )2 ⎟.
⎜ r US − r EU + (βUSX ΛUS ) ⎟
⎝ t 3⎠

3 | PORT FOLI O P LA NNING

The previous section established the international setup with stochastic variances and stochastic covariances. In the
next step, we study the impact of these additional sources of risk on international portfolio choice and subsequently on
the (different) benefits of international investors that result from international stock market participation. In
Section 3.1, we derive the wealth dynamics from the perspective of both investors. Section 3.2 solves the portfolio
planning problem of the domestic and foreign investor where we assume constant relative risk aversion.

3.1 | Dynamics of wealth


US and European investors can trade in international stocks and (locally risk-free) bonds. The trading strategy ωtc ∈ 3,
c ∈ {US , EU } gives the portfolio weights of US and European investors in international stocks and foreign bonds. More
precisely, ω3US denotes the portfolio weight the US investor assigns to European bonds, while ω3EU gives the weight
European investors assign to US bonds. The first two entries are the portfolio weights of the US stock and the EU stock.
The portfolio weights of the locally risk-free bonds are 1 − 1′ωtc . ΠUS EU
t and Πt denote the wealth of the US and the EU
investor in USD and EUR, respectively.

Lemma 3 The dynamics of wealth Πc are given by

dΠtc
= [r c + ωtc′ β cXt Λc]dt + ωtc ′β c Xt dZt (21)
Πtc

where c ∈ {US , EU }.
Assume that US and European investors pick the same exposures ω̄t = β EU ′ωtEU = βUS′ωtUS to Xt dZt . Then it
holds that
BRANGER ET AL. | 135

dΠUSt dΠtEU
− = [r US − r EU + (ω̄t ) ′Xt (ΛUS − ΛEU )] dt
ΠUS
t ΠtEU
(22)
= [r US − r EU + ωtUS ′YtUS δ ] dt
= [r EU − r US − ωtEU ′YtEU δ ] dt .

The difference of expected returns depends on (i) the interest rate differential, (ii) the covariances between stock returns
and the exchange rate Ytc,13 and Ytc,23, and (iii) the variance of the exchange rate Ytc,33.

3.2 | Portfolio planning problem


We assume that the investors have constant relative risk aversion (CRRA) and maximize the expected utility from
terminal wealth for a given investment horizon T > 0. The optimization problem is given by

c 1−γ
max E ⎡ (ΠT ) ⎤ c ∈ {US , EU }, (23)
ωtc ,0 ≤ t ≤ T ⎢⎣ 1−γ ⎥ ⎦

with relative risk aversion 0 < γ ≠ 1.9 The solution to the investment problem and the optimal portfolio strategy are
summarized in the next proposition.

Proposition 1 The indirect utility function of the investor is given by

(Πc )1 − γ
 c (t , Πc , X ) = exp{tr [Ac (τ ) Xt ] + C c (τ )}. (24)
1−γ

The optimal portfolio weights follow from

1 c
ω ct̃ = β c′ωtc = (Λ + 2Ac (τ ) Q′ρ), (25)
γ

where τ = T − t . The functions Ac and C c solve the system of ordinary differential equations

∂Ac (τ ) 1−γ
= Ac (τ ) Γ c + Γ c′Ac (τ ) + 2Ac (τ ) Q′ ⎜⎛ + ρρ′⎟⎞ QAc (τ ) + ζ c (26)
∂τ ⎝ γ ⎠

∂C c (τ )
= (1 − γ ) r c + tr [ΩΩ′Ac (τ )] (27)
∂τ

subject to the terminal conditions Ac (0) = 0, C c (0) = 0 and

1−γ
Γc = M + Q′ρ Λc′
γ
1−γ c c
ζc = Λ Λ ′.

 denotes the 3 × 3 identity matrix.

Proof Optimization with respect to the exposures ω c̃ = β c′ωc is a special case of the portfolio optimization
problem studied in Branger et al. (2017). A proof can be found in that paper and the references cited therein. □

9
We do not consider the special case γ = 1 representing the log‐investor. An optimization problem with a WASC model with logarithmic utility is treated by Bäuerle & Li (2013).
136 | BRANGER ET AL.

The time dependent functions Ac are provided in Appendix A and C c (τ ) follows by direct integration of
Equation (27). The functions Ac measure the sensitivity of the indirect utility function with respect to the state matrix
X . The more different AUS and AEU , the more different the impact of changes in the state matrix on the utility. Note that
the indirect utility depends on ΛUS and ΛEU , respectively, which capture the market prices of Z ‐risk. The risk premia for
variance‐covariance risk have no impact on the solution of the investment problem and on the optimal utility, since the
investors cannot trade variance and covariance risk due to market incompleteness.
Proposition 1 gives the optimal portfolios of the European and the US investor

1 c −1 c
ωtc = (β ′) (Λ + 2Ac (τ ) Q′ρ). (28)
γ

The optimal portfolio comprises a myopic (speculative) demand and a hedging demand. The myopic demand is driven
by the risk premia Λc of the tradeable assets. Given the definitions of βUS , β EU , ΛUS , and ΛEU , the myopic demand of
domestic and foreign investors for stocks is identical.
The hedging demand is

2 c −1 c 2 2
ωtc, hedge = (β ′) A (τ ) Q′ρ = (β c′)−1Ac (τ )(β c )−1β cQ′ρ = A*c (τ ) Q *c′ρ , (29)
γ γ γ

where we define

A*c = (β c′)−1Ac (τ )(β c )−1


(30)
Q *c = Qβ c′.

It is driven by the desire of the investor to hedge unanticipated changes of the variance‐covariance matrix Y c over the
investment horizon. Note that it holds that

tr[Ac (τ ) Xt ] = tr[Ac (τ )(β c )−1β cXt β c′ (β c′)−1]


= tr[(β c′)−1Ac (τ )(β c )−1β cXt β c′] = tr[A*c (τ ) Ytc ].

Hence, the hedging demand depends on the impact A*c of Y c on the indirect utility function  in Equation (24) as well
as on the covariance between the returns of (traded) assets and (non‐traded) variance‐covariance risk, which is driven
by Q *c . We follow Buraschi et al. (2010) and decompose the hedging demand into

*c (τ )(Q *c ′ρ)1 ⎞
⎛ A11 *c (τ )(Q *c ′ρ)2 + A13
⎛ A12 *c (τ )(Q *c ′ρ)3 ⎞
2⎜ c 2
ωtc, hedge = A * (τ )(Q *c ′ρ)2 ⎟ + ⎜ A12 *c (τ )(Q *c ′ρ)3 ⎟.
*c (τ )(Q *c ′ρ)1 + A23 (31)
γ ⎜ 22 ⎟ γ⎜ ⎟
*c c′ *c c′ *c c′
⎝ A33 (τ )(Q * ρ)3 ⎠ ⎝ A13 (τ )(Q * ρ)1 + A23 (τ )(Q * ρ)2 ⎠

The first vector summarizes the hedging demand due to changes in variances while the second vector is the hedging
demand due to changes in covariances ( Aii*c and Aij*c , i ≠ j are the sensitivities with respect to variances Yiic and
covariances Yijc , respectively). The hedging demand vanishes for (Q *c′ρ) = 0.

4 | ASSET ALLO C A TI O N O N IN C O M P L E T E MAR K ET S

In this section we provide a numerical example which highlights the importance of variance‐covariance risk for optimal
portfolios and investors’ utility. In Section 4.1 we calibrate the model using data on the S&P 500 index, the Euro Stoxx
50 index, and the USD/EUR exchange rate. Section 4.2 contains the numerical results.
BRANGER ET AL. | 137

T A B L E 1 Base case parameters

Interest rates

r US 0.00
r EU 0.00
State variables
X0 M
⎛ 0.0282 0.0173 0.0033 ⎞ ⎛−2.7500 0.1000 0.0100 ⎞
⎜ 0.0173 0.0297 0.0092 ⎟ ⎜ 0.1000 −2.7500 0.0100 ⎟
⎝ 0.0033 0.0092 0.0120 ⎠ ⎝ 0.0100 0.0100 −4.0000 ⎠
Q ρ
⎛ 0.2500 0.0800 0.0100 ⎞ ⎛−0.5000 ⎞
⎜ 0.0800 0.2500 0.0600 ⎟ ⎜−0.4000 ⎟
⎝ 0.0100 0.0600 0.2000 ⎠ ⎝ 0.2500 ⎠
k 2.2000
Market prices of risk
ΛUS ⎛ 0.9963 ⎞ ΛEU ⎛ 0.9963 ⎞
⎜ 1.3510 ⎟ ⎜ 1.3510 ⎟
⎝−0.4701⎠ ⎝−1.4701⎠
Note. The table gives the base case parametrization for our model. The matrix X0 is set to its long‐run mean.

4.1 | Parametrization and set‐up


To calibrate the model, we rely on daily index returns of the S&P 500, the Euro Stoxx 50, and the USD/EUR exchange
rate. While both stock market indices are price indices, we consider returns that are adjusted for dividend payments.
The observation period is January 2002 to December 2015. Data was obtained from Thomson Reuters Eikon.
The parameters for our base case are given in Table 1. They are in line with the relevant empirical literature. For an
extensive overview we refer to Branger et al. (2017) and the references cited therein. In particular, we set the long‐run
average X∞ equal to the variance‐covariance matrix of returns (denominated in USD). Furthermore, βUS is the identity
matrix, that is, Xt measures the variance‐covariance matrix of USD‐denominated returns. As Equation (3) shows, the
choice of X∞ is related to k , M and Q . Volatility smiles are downward sloping for stock market indices and symmetric
for exchange rates. The correlation between stock returns and their variances (leverage) is thus negative, while this
correlation is around 0 for exchange rate returns.10 The parameters ρ and Q are chosen to obtain reasonable levels for
these correlations. To ensure that the variance‐covariance matrix X is positive semi‐definite, we set k = 2.2.

T A B L E 2 Moments and risk premia

US perspective EU perspective
Long run second moments
long‐run volatilities
⎛ 0.1681 ⎞ ⎛ 0.1838 ⎞
⎜ 0.1722 ⎟ ⎜ 0.1527 ⎟
⎝ 0.1098 ⎠ ⎝ 0.1098 ⎠
long‐run correlations
⎛ 1 0.5982 0.1763 ⎞ ⎛ 1 0.6026 0.4360 ⎞
⎜ 0.5982 1 0.4865 ⎟ ⎜ 0.6026 1 0.1702 ⎟
⎝ 0.1763 0.4865 1 ⎠ ⎝ 0.4360 0.1702 1 ⎠
leverage
⎛−0.5882 ⎞ ⎛−0.5721 ⎞
⎜−0.4642 ⎟ ⎜−0.5931 ⎟
⎝ 0.1005 ⎠ ⎝−0.1005 ⎠
Risk premia
excess returns
⎛ 0.0500 ⎞ ⎛ 0.0488 ⎞
⎜ 0.0530 ⎟ ⎜ 0.0459 ⎟
⎝ 0.0100 ⎠ ⎝ 0.0020 ⎠
Note. The table gives the long run second moments and risk premia as seen from the US and EU perspective. The first and second row of the vectors and
matrices refer to US and EU stocks, while the third row refers to the foreign bond (EU and US bond from the US and the EU perspective, respectively). The first
panel provides return volatilities, return correlations and the correlation between returns and their variances (leverage). Leverages are calculated along the lines
of Equation (7). The second panel gives the expected excess returns on the assets.
138 | BRANGER ET AL.

(a)

(b)

F I G U R E 1 Optimal portfolio holdings. The figure shows optimal portfolio weights for the US investor (left column) and the European
investor (right column). Hedging demands are calculated as percentage of the myopic demand, both for the US investor (left column)
and the European investor (right column). Unless stated otherwise, we set γ = 5 and τ = 1 year [Color figure can be viewed at
wileyonlinelibrary.com]

Table 2 summarizes the resulting moments. It also contains the stock and bond risk premia which are exogenously
specified for the USD‐returns. The premia from the perspective of the EU investor follow from the European pricing
kernel. Risk premia are chosen to lie in a generally accepted range. Moreover, we set interest rates (and thus the interest
rate differential) equal to zero.
To analyze the economic benefits from trading, we rely on certainty equivalent returns (CER) which are defined as

(ΠTc )1 −γ ⎤
u (Πc0 e CERT T ) = E ⎡
c

⎢ 1 − γ ⎥,
⎣ ⎦

where c ∈ {US , EU }. They give the return per year for which an investor is indifferent between receiving this
(deterministic) return and the (stochastic) terminal wealth of the risky trading strategy.11 Since interest rates are assumed to

10
For a discussion on the impact of correlation between stock returns and variances (leverage) and the slope of the volatility smile see Heston (1993) for the single stock case and, for example, Branger
& Muck (2012) for a model with Wishart processes.
11
This benchmark measure is thus closely related to the certainty equivalent wealth which is, for example, considered by Liu & Pan (2003).
BRANGER ET AL. | 139

be zero, the CER of a domestic bond‐only investor is zero and the CER can be interpreted as the utility gain from trading in
the risky assets.12 Both investors are assumed to have identical risk aversion (γ = 5) and a 1‐year investment horizon.

4.2 | Numerical results


Figure 1 gives the optimal portfolios. Panel A shows the optimal portfolio weights of the US investor (left graph) and the
EU investor (right graph). The optimal allocations to the risky stocks are nearly identical for both investors. They
allocate approximately 21% and 29% of their wealth to the US stock and the EU stock, respectively. In contrast, the
optimal holdings of the investors differ significantly when it comes to the foreign bond investments. While the US
investor borrows 10% of his wealth in foreign currency (EUR), the corresponding position of the European investor is
nearly twice as large (19% in USD). Both investors thus enter short positions even though the risk premium on foreign
bonds is positive. This can be explained by their desire to hedge foreign exchange rate risk. Foreign bonds are highly
correlated with the foreign stocks due to their (joint) exposure to exchange rate risk. A short position in foreign bonds
reduces the overall exposure to the exchange rate. The short position due to this hedging motive is larger than the long
position induced by the positive premium on the bond, which results in an overall short position in the foreign bond.
The difference in the overall position in foreign bonds reflects the different risk premia faced by both investors.
Next, we analyze the hedging demands in more detail. Hedging demands arise due to the stochastic variance‐
covariance matrices Y US = X and Y EU . Panel B of Figure 1 gives the variance and the covariance hedging demand
obtained from Equation (31) as a percentage of the myopic demand for the US and the European investor. The variance
hedging demand in the stocks is around 3% of the myopic demand for both investors, and is slightly larger for the
European investor. Thus, the size of the hedging demand seems to be in a plausible range found by other authors as
well.13 The covariance hedging demands in the stocks are of comparable size (up to 3%) for both investors.
Again, the major difference between both investors’ hedging demands is observed in the foreign bond holdings. For
the European investor the hedging demand is about twice as important as for her US counterpart. This difference is
mainly driven by the covariance hedging demand. Technically, these results can be traced back to the differences in the
investment opportunity sets as seen by the US and the European investor and the differences in the dynamics of second
moments. Furthermore, Equation (26) shows that the differences in the expected returns on wealth are driven by the
variance of the exchange rate and its covariances with the other assets. As argued above, this implies different myopic
positions in the foreign bonds, and it also implies a different exposure to these second moments, which then results in a
different hedging demand in the foreign bond.
Figure 2 shows the relationship between the CERs (which measure the benefit from trading) and the
correlations between the assets. The first row gives the dependence on ρ̄12 , that is, on the correlation between the
stock returns. We vary the covariance X12 = X21 while holding the variances X11 and X22 constant. Both CER
increase in the correlation (covariance). Here, the benefits from the larger risk premia on stocks (note that both the
first and second entry in ΛUS and ΛEU are positive) more than outweigh the utility loss due to less diversification.
Independent of the correlation, the difference of the certainty equivalent returns remains constant, and it is always
the US investor who has the larger CER.
The middle and lower row of Figure 2 show the impact of the correlations (covariances) between the exchange rate and
the stocks. Here, we vary the covariances X13 = X31 and X23 = X32 and keep all other variances constant. The correlations
affect the investment opportunity sets of the foreign and domestic investor differently. Consequently, the changes in the CER
also differ from each other. For both investors higher correlations lead to lower certainty equivalent returns. However,
European investors are more strongly affected. When correlations increase, their CER drops more than that of the US
investor. As seen in the lower row, European investors might even have a higher CER than US investors when the correlation
between the European stock and the exchange rate is low and a lower one when the correlation is high.

5 | ASSET ALLO C A TI O N I N CO M P L ET E M A R K ET S

As it is, for example, highlighted by Liu & Pan (2003), the benefits from trading can improve substantially when investors also
have access to the derivatives market. The prices of financial derivatives in general depend on stock variances and covariances.

12
As noted above, differences between certainty equivalent returns are not affected by the choice of (deterministic) interest rates.
13
An overview is provided, for example, in Branger et al. (2017). As compared to Buraschi et al. (2010), however, the hedging demands reported in Figure 1 are small.
140 | BRANGER ET AL.

F I G U R E 2 Certainty equivalent returns and correlation. The figure shows the certainty equivalent returns (CER) as functions of local
return correlations. The solid lines give the US investor’s CER, the dashed lines refer to the CER of the European investor. The investment
horizon is τ = 1 year, the relative risk aversion is γ = 5 [Color figure can be viewed at wileyonlinelibrary.com]

Hence, they can be used to get an exposure to the state matrix X . This allows investors to disentangle stock market risk from
variance‐covariance risk. They can then also earn the premia on variance and covariance risk, and they can hedge the risk in
second moments better, which leads to higher certainty equivalent returns.
The purpose of this section is to analyze the impact of stochastic covariances and correlations on domestic and
foreign investors when markets are complete. In Section 5.1 we compare their investment opportunity sets and state the
optimal portfolios. In Section 5.2 we present a numerical example.

5.1 | Portfolio planning


On a complete market an investor can attain arbitrary exposures θ B, c ∈ n × n and θW , c ∈ n to the risk factors B and W .
These exposures are linear combinations of the portfolio weights in risky assets and derivatives. The following lemma
gives the dynamics of wealth depending on the home currency of the investor.

Lemma 4 The dynamics of wealth Πc are given by

dΠtc
= r c dt + tr [(θtB, c ) ′ (Xt ηc dt + Xt dBt )] + (θtW , c ) ′ (Xt ξ c dt + Xt dWt ). (32)
Πtc
BRANGER ET AL. | 141

Proof For the given exposures θtB, c and θtW , c the expected excess returns over the risk free rate of interest follow
directly from the pricing kernels (10) and (16). For further details, see also, for example, Branger et al. (2017). □
B W
Assume that US and EU investors pick the same exposures θt̄ = θtB, US = θtB, EU and θt̄ = θtW , US = θtW , EU to the risk
factors. The difference between the returns from the perspective of a European and a US investor is

dΠUS dΠtEU B
t
US
− EU
= {r US − r EU + tr [(θt̄ ) ′Xt (ηUS − η EU )] + (θW
̄ )′Xt (ξ US − ξ EU )} dt
Πt Πt (33)
= {
r US − r EU + tr [θtB, US′ ( )
βUS −1YtUS δρ′ ]+ θtW , US′ ( )
βUS −1YtUS δ 1 − ρ′ρ } dt

As in the incomplete market case, investors earn different expected returns for identical factor exposures. The
difference depends on (i) the interest rate differential, (ii) the covariances between stocks and the exchange rate (YtUS
,13
and YtUS US
,23), and (iii) the variance of the exchange rate (Yt ,33).
As in the incomplete market, we assume that the US and the European investor have identical CRRA preferences
and maximize the expected utility from terminal wealth. The solution to the investment problem is summarized in the
next Proposition.

Proposition 2 The indirect utility function of the investor is given by

(Πc )1 − γ
 c (t , Πc , X ) = exp {tr ( c (τ ) Xt ) +  c (τ )},
1−γ

and the optimal exposures are

1 c
θtB, c = (η + 2 c (τ ) Q′)
γ
1
θtW , c = ξc
γ

where τ = T − t . The functions  c and  c solve the system

∂ c (τ ) 2
=  c (τ ) Γ c + Γ c′ c (τ ) +  c (τ ) Q′Q  c (τ ) + ζ c (34)
∂τ γ

∂ c (τ )
= (1 − γ ) r c + tr [ΩΩ′ c (τ )] (35)
∂τ

subject to the terminal conditions  c (0) = 0,  c (0) = 0 and

1−γ
Γc = M + Q′ηc ′
γ
1−γ c c
ζc = [η η ′ + ξ cξ c′].

Proof This is a special case of the portfolio optimization problem treated in Branger et al. (2017). A proof can be
found in that paper and the references cited therein. □

The solutions of Equations 34 and 35 are stated in Appendix A. As in the incomplete market, the optimal demand
comprises myopic and hedging components. The constant myopic demand follows from the market prices of risk for dB
and dW . The hedging demand depends on the matrix  c (τ ), which denotes the sensitivity of  to changes in X , and on
the matrix Q , which captures the exposure of X to dB .
142 | BRANGER ET AL.

T A B L E 3 Parameters and risk premia for the complete market

US perspective EU perspective
Market prices of risk
ηUS ⎛−0.7847 0.0557 0.1447 ⎞ η EU ⎛−0.7847 0.0557 0.1447 ⎞
⎜−0.2309 −0.7429 −0.0876 ⎟ ⎜−0.2309 −0.7429 −0.0876 ⎟
⎝−0.0227 −0.2442 −0.0240 ⎠ ⎝ 0.4773 0.1558 −0.2740 ⎠
ξ US ⎛ 0.8125 ⎞ ξ EU ⎛ 0.8125 ⎞
⎜ 1.3220 ⎟ ⎜ 1.3220 ⎟
⎝−0.7891⎠ ⎝−1.5154 ⎠
Variance‐covariance risk
premia
1 1
⎛−0.0150 −0.0120 −0.0025 ⎞ ⎛−0.0146 −0.0083 −0.0013 ⎞
US EU
(Et [dYtUS] − Et [dYtUS]) (Et [dYtEU ] − Et [dYtEU ])
dt ⎜−0.0120 −0.0150 −0.0045 ⎟ dt ⎜−0.0083 −0.0083 0.0012 ⎟
⎝−0.0025 −0.0045 −0.0015 ⎠ ⎝−0.0013 0.0012 −0.0020 ⎠
Note. The table gives the additional parameters and risk premia as seen from the US and EU perspective for the complete market analysis.

Proposition 2 gives the optimal exposures to the risk factors B and W . We are, however, mainly interested in the
positions in stocks and bonds and in the positions in variance and covariance risk. Therefore, we follow Branger et al.
(2017) and re‐write the wealth dynamics given in Lemma 4 as

dΠtc
= ΘV , c ′diag(Vtc )−1dVtc + tr [ΘY , c ′dYtc ], (36)
Πtc

where the optimal exposures are14

1 1
ΘVt , c = (β c′)−1ξ c (37)
γ 1 − ρ′ρ

1 c −1 ⎡ c 1 ⎤ 1
ΘYt , c = (β ′) ⎢η − ξ cρ′⎥ (Q′)−1 (β c )−1 + (β c′)−1 c (τ )(β c )−1. (38)
2γ 1 − ρ′ρ γ
⎣ ⎦

The demand for the assets is driven by the market prices of risk and coincides with the myopic demand. The demand
for variance‐covariance risk has both a myopic component and a hedging component, where the latter depends on  c .

5.2 | Numerical results


For the numerical example we again consider US and European investors with identical preferences, risk aversion, and
planning horizons. We rely on the base case parametrization given in Table 1. Again, βUS is the identity matrix. In
addition, the (long‐run) risk premia for variance‐covariance risk must be specified. Details are provided in Table 3. The
risk premia meet technical conditions15 and imply negative variance‐covariance risk premia which are in line with the
literature.16 The market prices of risk and the risk premium for variance‐covariance risk as seen by the European
investor follow endogenously. Note that the level of ηUS and ξ US affects the level of the investors’ utility, but not the
difference in their CERs.
Panel A of Figure 3 compares the exposures to variance and covariance risk for the US and the European
investor according to Equation (38). We find that there are significant differences for the exposures to the
covariances between the exchange rate and the stock returns as well as to the variance of the exchange rate. In

14
As in Branger et al. (2017), the coefficients ηc and ξ c have to meet additional conditions to ensure that ΘY , c is symmetric. They can always be chosen such that these conditions hold true. The reason
is that, given the premia on stocks, bonds, and second moments, the coefficients ηc and ξ c are not unique. While we can basically use any solution for ηc and ξ c which results in the given risk premia,
there is one specific solution which is needed for Equations 37 and 38 to hold.
15
See Footnote 14.
16
Pan (2002), Bakshi & Kapadia (2003a,2003b), Carr & Wu (2009) or, for international markets, Londono & Zhou (2017) consider negative variance risk premia. Krishnan et al. (2009), Driessen et al.
(2012), Mueller, Stathopoulos and Vedolin (2017) and Hollstein & Simen (2017) report significantly negative compensations for correlation risk.
BRANGER ET AL. | 143

(a)

(b)

F I G U R E 3 Optimal exposures and certainty equivalent returns (complete market). The figure shows the optimal exposure ΘY , US and
ΘY , EU and the certainty equivalent returns (CER) as a function of the investment horizon τ and local correlations, respectively. The solid
lines refer to the US investor, the dashed lines give the exposure of the European investor. The investment horizon is τ = 1 year and the
coefficient of relative risk aversion is γ = 5 [Color figure can be viewed at wileyonlinelibrary.com]

particular, while the US investor has negative exposures to the covariances of the foreign bond with the stocks and
a positive exposure to the variance of the foreign bond, the opposite holds true for the European investor.
Technically, these observations are in line with Equation (38) which highlights that for identical exposures
investors face different expected returns depending on the variance of the exchange rate as well as covariances
between stocks and the exchange rate. We also look at the exposures to the second moments of stock returns. They
are nearly identical for the domestic and the foreign investor. This is demonstrated for the example of the
correlation between stock returns in Figure 3.
Panel B of Figure 3 gives the certainty equivalent returns when the state matrix X differs from its long term mean.
Local correlations affect the CER in a similar way as in the incomplete market. Again, the CER of both investors
increases in the stock market correlation, while the difference of the CERs stays constant. Furthermore, higher
correlations between stocks and the exchange rate lower the CER of both investors. The size of the impact differs
between the investors in that European investors suffer larger losses than US investors. Similar to the incomplete
market, European investors might even have higher CERs when the correlation between the European stock and the
exchange rate is low, while the opposite holds true when the correlation is high.
144 | BRANGER ET AL.

6 | C ON C LU S I O N

In this paper, we study the optimal asset allocation from an international perspective when variances, covariances, and
correlations between stocks and the exchange rate are stochastic. We derive optimal portfolio strategies from the
domestic and the foreign perspective and study the differences in investors’ benefits from international trading.
We find that correlation risk matters to investors and that the impact of exchange rate risk on domestic and foreign
investors is not necessarily the same. First, positions in foreign bonds differ between investors while their holdings of
stocks are very similar. Second, hedging demands in foreign bonds are different for the investors in the incomplete
market. In the complete market, positions in the variance of the exchange rate and in the covariances between stock
returns and the exchange rate also differ significantly for both investors. Third, benefits from international trading
depend on stochastic correlations. While both investors react in the same way to changes in the correlation between
risky stocks, changes in the correlation between the exchange rate and stock returns again affect the investors
differently and may even change the relative size of the utility gains from trading.
We base our analysis on exogenously specified market prices of risk. An issue for future research is to study the
pricing of variance‐covariance risk in an international equilibrium model. It would also be interesting to apply the
model to the pricing of stock and foreign exchange options which then allows to extract the risk‐neutral dynamics and
thus also the compensations on variance‐covariance risk. Finally, the inclusion of jump risk, as proposed by Leippold &
Trojani (2010), allows to study the different reactions of investors in particular to sudden changes of second moments.

ORCID
Nicole Branger http://orcid.org/0000-0003-0466-6515
Matthias Muck http://orcid.org/0000-0003-2364-9833

REFERENCES
Adler, M., & Dumas, B. (1983). International portfolio choice and corporation finance: A synthesis. Journal of Finance, 38(3), 925–984.
https://doi.org/10.1111/j.1540‐6261.1983.tb02511.x
Ang, A., & Bekaert, G. (2002). International asset allocation with regime shifts. Review of Financial Studies, 15(4), 1137–1187. https://doi.org/
10.1093/rfs/15.4.1137
Bäuerle, N., & Li, Z. (2013). Optimal portfolios for financial markets with Wishart volatility. Journal of Applied Probability, 1025–1043.
https://doi.org/10.1239/jap/1389370097
Bakshi, G., & Kapadia, N. (2003a). Delta‐Hedged gains and the negative market volatility risk premium. Review of Financial Studies, 16(2),
527–566. https://doi.org/10.1093/rfs/hhg002
Bakshi, G., & Kapadia, N. (2003b). Volatility risk premiums embedded in individual equity options: Some new insights. Journal of
Derivatives, 11(1), 45–54. https://doi.org/10.3905/jod.2003.319210
Ball, C. A., & Torous, W. N. (2000). Stochastic correlation across international stock markets. Journal of Empirical Finance, 7(3‐4), 373–388.
https://doi.org/10.1016/S0927‐5398(00)00017‐7
Bekaert, G., & Urias, M. S. (1996). Diversification, integration and emerging market closed‐end funds. Journal of Finance, 51(3), 835. https://
doi.org/10.1111/j.1540‐6261.1996.tb02709.x
Bollerslev, T., Engle, R. F., & Wooldridge, J. M. (1988). A capital asset pricing model with time‐varying covariances. Journal of Political
Economy, 96(1), 116–131. https://doi.org/10.1086/261527
Branger, N., & Muck, M. (2012). Keep on smiling? The pricing of Quanto options when all covariances are stochastic. Journal of Banking &
Finance, 36(6), 1577–1591. https://doi.org/10.1016/j.jbankfin.2012.01.004
Branger, N., Schlag, C., & Schneider, E. (2008). Optimal portfolios when volatility can jump. Journal of Banking & Finance, 32(6), 1087–1097.
https://doi.org/10.1016/j.jbankfin.2007.09.015
Branger, N., Muck, M., Seifried, F. T., & Weisheit, S. (2017). Optimal portfolios when variances and covariances can jump. Journal of
Economic Dynamics and Control, 85, 59–89. https://doi.org/10.1016/j.jedc.2017.09.008
Brennan, M. J., & Xia, Y. (2002). Dynamic asset allocation under inflation. Journal of Finance, 57(3), 1201–1238. https://doi.org/10.1111/
1540‐6261.00459
Bru, M.‐F. (1991). Wishart processes. Journal of Theoretical Probability, 4(4), 725–751. https://doi.org/10.1007/BF01259552
Buraschi, A., Porchia, P., & Trojani, F. (2010). Correlation risk and optimal portfolio choice. Journal of Finance, 65(1), 393–420.
https://doi.org/10.1111/j.1540‐6261.2009.01533.x
Buss, A., Schoenleber, L. & Vilkov, G. (2018). Expected stock returns and the correlation risk premium (Working paper).
Carr, P., & Wu, L. (2009). Variance risk premiums. Review of Financial Studies, 22(3), 1311–1341. https://doi.org/10.1093/rfs/hhn038
BRANGER ET AL. | 145

Chacko, G., & Viceira, L. M. (2005). Dynamic consumption and portfolio choice with stochastic volatility in incomplete markets. Review of
Financial Studies, 18(4), 1369–1402. https://doi.org/10.1093/rfs/hhi035
Chiarella, C., Hsiao, C.‐Y. & To, T. D. (2011). Stochastic correlation and risk premia in term structure models (Working paper).
Cox, J. C., Ingersoll, J. E., & Ross, S. A. (1985). A theory of the term structure of interest rates. Econometrica, 53(2), 385–407. https://doi.org/
10.2307/1911242
Cuchiero, C., Filipović, D., Mayerhofer, E., & Teichmann, J. (2011). Affine processes on positive semidefinite matrices. Annals of Applied
Probability, 21(2), 397–463. https://doi.org/10.1214/10‐AAP710
Da Fonseca, J., Grasselli, M., & Tebaldi, C. (2007). Option pricing when correlations are stochastic: An analytical framework. Review of
Derivatives Research, 10(2), 151–180. https://doi.org/10.1007/s11147‐008‐9018‐x
Da Fonseca, J., Grasselli, M., & Ielpo, F. (2011). Hedging (Co) variance risk with variance swaps. International Journal of Theoretical &
Applied Finance, 14(2), 899–943. https://doi.org/10.1142/S0219024911006784
De Col, A., Gnoatto, A., & Grasselli, M. (2013). Smiles all around: FX joint calibration in a multi‐heston model. Journal of Banking & Finance,
37(10), 3799–3818. https://doi.org/10.1016/j.jbankfin.2013.05.031
Driessen, J., Maenhout, P. J., & Vilkov, G. (2009). The price of correlation risk: Evidence from equity options. Journal of Finance, 64(3), 1377–
1406. https://doi.org/10.1111/j.1540‐6261.2009.01467.x
Driessen, J., Maenhout, P. J., & Vilkov, G. (2012). Option‐implied correlations and the price of correlation risk (Working paper).
Egloff, D., Leippold, M., & Wu, L. (2010). The term structure of variance swap rates and optimal variance swap investments. Journal of
Financial and Quantitative Analysis, 45(05), 1279–1310. https://doi.org/10.1017/S0022109010000463
Engle, R. (2002). Dynamic conditional correlation: A simple class of multivariate generalized autoregressive conditional heteroskedasticity
models. Journal of Business & Economic Statistics, 20(3), 339–350. https://doi.org/10.1198/073500102288618487
Escobar, M., Ferrando, S. & Rubtsov, A. (2015). International portfolio choice under multi‐factor stochastic volatility (Working paper).
Goetzmann, W. N., Li, L., & Rouwenhorst, K. G. (2005). Long‐term global market correlations. Journal of Business, 78(1), 1–38.
https://doi.org/10.1086/426518
Gourieroux, C. (2006). Continuous time wishart process for stochastic risk. Econometric Reviews, 25(2‐3), 177–217. https://doi.org/10.1080/
07474930600713234
Gruber, P. H., Tebaldi, C., & Trojani, F. (2014). The price of the smile (Working paper).
Heston, S. L. (1993). A closed‐form solution for options with stochastic volatility with applications to bond and currency options. Review of
Financial Studies, 6(2), 327–343. https://doi.org/10.1093/rfs/6.2.327
Hollstein, F., & Simen, C. W. (2017). Variance risk: A bird’s eye view (Working paper).
Kaplanis, E., & Schaefer, S. M. (1991). Exchange risk and international diversification in bond and equity portfolios. Journal of Economics
and Business, 43(4), 287–307. https://doi.org/10.1016/0148‐6195(91)90027‐T
Krishnan, C. N. V., Petkova, R., & Ritchken, P. (2009). Correlation risk. Journal of Empirical Finance, 16(3), 353–367. https://doi.org/10.1016/
j.jempfin.2008.10.005
Larsen, L. S. (2010). Optimal investment strategies in an international economy with stochastic interest rates. International Review of
Economics and Finance, 19(1), 145–165. https://doi.org/10.1016/j.iref.2009.04.001
Leippold, M. & Trojani, F. (2010). Asset pricing with matrix jump diffusions (Working paper).
Li, K., Sarkar, A., & Wang, Z. (2003). Diversification benefits of emerging markets subject to portfolio constraints. Journal of Empirical
Finance, 10(1‐2), 57–80. https://doi.org/10.1016/S0927‐5398(02)00027‐0
Lioui, A., & Poncet, P. (2003). International asset allocation: A new perspective. Journal of Banking & Finance, 27(11), 2203–2230.
https://doi.org/10.1016/S0378-4266(02)00325-4
Liu, J., & Pan, J. (2003). Dynamic derivative strategies. Journal of Financial Economics, 69(3), 401–430. https://doi.org/10.1016/S0304‐405X
(03)00118‐1
Liu, J., Longstaff, F. A., & Pan, J. (2003). Dynamic asset allocation with event risk. Journal of Finance, 58(1), 231–259. https://doi.org/
10.1111/1540‐6261.00523
Liu, J. (2007). Portfolio selection in stochastic environments. Review of Financial Studies, 20(1), 1–39. https://doi.org/10.1093/rfs/hhl001
Londono, J. M., & Zhou, H. (2017). Variance risk premiums and the forward premium puzzle. Journal of Financial Economics, 124(2),
415–440. https://doi.org/10.1016/j.jfineco.2017.02.002
Longin, F., & Solnik, B. (1995). Is the correlation in international equity returns constant: 1960‐1990? Journal of International Money and
Finance, 14(1), 3–26. https://doi.org/10.1016/0261‐5606(94)00001‐H
Merton, R. C. (1969). Lifetime portfolio selection under uncertainty: The continuous‐time case. Review of Economics and Statistics, 51(3),
247–257. https://doi.org/10.2307/1926560
Merton, R. C. (1971). Optimum consumption and portfolio rules in a continuous‐time model. Journal of Economic Theory, 3(4), 373–413.
https://doi.org/10.1016/0022‐0531(71)90038‐X
Merton, R. C. (1973). An intertemporal capital asset pricing model. Econometrica, 41(5), 867–887. https://doi.org/10.2307/1913811
Muck, M. (2010). Trading strategies with partial access to the derivatives market. Journal of Banking & Finance, 34(6), 1288–1298. https://doi.org/
10.1016/j.jbankfin.2009.11.025
Mueller, P., Stathopoulos, A., & Vedolin, A. (2017). International correlation risk. Journal of Financial Economics, 126(2), 270–299.
https://doi.org/10.1016/j.jfineco.2016.09.012
146 | BRANGER ET AL.

Munk, C., Sørensen, C., & Nygaard Vinther, T. (2004). Dynamic asset allocation under mean‐reverting returns, stochastic interest rates, and
inflation uncertainty: Are popular recommendations consistent with rational behavior? International Review of Economics and Finance,
13(2), 141–166. https://doi.org/10.1016/j.iref.2003.08.001
Pan, J. (2002). The jump‐risk premia implicit in options: Evidence from an integrated time‐series study. Journal of Financial Economics, 63
(1), 3–50. https://doi.org/10.1016/S0304‐405X(01)00088‐5
Smedts, K. (2004). International dynamic asset allocation and the effect of the exchange rate (Working paper).

How to cite this article: Branger N, Muck M, Weisheit S. Correlation risk and international portfolio choice.
J Futures Markets. 2019;39:128–146. https://doi.org/10.1002/fut.21941

APPENDIX A

Solution to equation (26)


The solution to Equation (26) is given by

Ac (τ ) = ℋ 22
c
(τ )−1ℋ 21
c
(τ ) (A.1)

where

c c


c
(
1−γ
⎛⎜ ℋ11 (τ ) ℋ12 (τ ) ⎞⎟ ≔ exp ⎢τ ⎛ Γ − 2Q′  + γ ρρ′ Q ⎞ ⎥.



)
⎝ℋ 21 (τ ) ℋ 22 (τ ) ⎠
c c
⎢ ⎜ζc − Γ c′ ⎟⎥
⎣ ⎝ ⎠⎦

Solutions to Equations 34 and 35


The solution to Equations 34 and 35 is given by

 c (τ ) = ℱ22
c
(τ )−1ℱ21
c
(τ ) (A.2)

γk
 c (τ ) = (1 − γ ) r cτ − tr [lnℱ22
c
(τ ) + Γ c′τ ] (A.3)
2

where

2
⎛⎜ ℱ11 (τ ) ℱ12 (τ ) ⎞⎟ ≔ exp ⎡τ ⎛ Γ − γ Q′Q ⎞ ⎤.
c c c
⎢ ⎥
⎝ ℱ21 (τ ) ℱ22 (τ ) ⎠
c c ⎜ c c′ ⎟

⎣ ⎝ ζ − Γ ⎠⎥⎦

Vous aimerez peut-être aussi