a discontinuous mispricing model under asymmetric information
TRANSCRIPT
European Journal of Operational Research 243 (2015) 944–955
Contents lists available at ScienceDirect
European Journal of Operational Research
journal homepage: www.elsevier.com/locate/ejor
Decision Support
A discontinuous mispricing model under asymmetric information
Winston S. Buckley a,∗, Hongwei Long b
a Department of Mathematical Sciences, Bentley University, Waltham, MA 02452, USAb Department of Mathematical Sciences, Florida Atlantic University, Boca Raton, FL 33431-0991, USA
a r t i c l e i n f o
Article history:
Received 23 October 2013
Accepted 25 December 2014
Available online 7 January 2015
Keywords:
Mispricing
Levy jumps
Asymmetric information
Optimal portfolio
Expected utility
a b s t r a c t
We study a discontinuous mispricing model of a risky asset under asymmetric information where jumps in
the asset price and mispricing are modelled by Lévy processes. By contracting the filtration of the informed
investor, we obtain optimal portfolios and maximum expected utilities for the informed and uninformed
investors. We also discuss their asymptotic properties, which can be estimated using the instantaneous
centralized moments of return. We find that optimal and asymptotic utilities are increased due to jumps in
mispricing for the uninformed investor but the informed investor still has excess utility, provided there is not
too little or too much mispricing.
© 2015 Elsevier B.V. All rights reserved.
T
i
e
c
i
a
t
t
w
w
a
a
a
p
w
1
m
f
t
f
p
1 The literature related to mean-reversion in asset prices is very large and includes
stocks: Poterba and Summers (1988), Depenya and Gil-Alana (2002), Gropp (2004),
Mukherji (2011, 2012); currencies: Engel and Hamilton (1990), Cheung and Lai (1994),
1. Introduction
Asymmetric information models assume that there are two types
of investors in the market: informed and uninformed. The informed
investors (e.g., institutional investors with internal research capabili-
ties) observe both fundamental and market prices, while uninformed
investors (e.g., retail investors who rely on public information in or-
der to make investment choices) observe market prices only. The
uninformed investors are viewed as liquidity traders or hedgers. The
prevalence of informed traders affects liquidity, transaction costs, and
trading volumes. The informed investors partially reveal information
through trades, which can cause higher permanent price changes.
Asymmetric information asset pricing models rely on a noisy rational
expectation equilibrium in which prices only partially reveal the in-
formed investor’s information. Many empirical studies confirm that
information asymmetry is priced and imply that liquidity is a primary
channel that links information asymmetry to prices (see, e.g., Admati,
1985; Easley & O’Hara, 2004; Grossman & Stiglitz, 1980; Hellwig,
1980; Kelly & Ljungqvist, 2012; Wang, 1993).
Mispricing is the difference between the asset’s market price and
fundamental value. The fundamental value can be defined as the
market price that would prevail if all the market participants were
perfectly informed investors. Because of the mean-reverting nature
of the mispricing process, it is typically modelled by a continuous
Ornstein–Uhlenbeck (O–U) process, while the price of the risky asset
is usually modelled by a continuous geometric Brownian motion (see,
e.g., Buckley, Brown, & Marshall, 2012; Guasoni, 2006; Wang, 1993).
∗ Corresponding author.
E-mail addresses: [email protected] (W. S. Buckley), [email protected]
(H. Long).
S
B
(
v
(
http://dx.doi.org/10.1016/j.ejor.2014.12.045
0377-2217/© 2015 Elsevier B.V. All rights reserved.
he mean-reverting speed or equivalently, the mean reverting-time,
s a proxy for mispricing. Mean-reversion1 is well-documented in the
mpirical financial literature and applies to asset returns, stock prices,
urrencies/exchange rates, interest rates, commodities, indexes, stock
ndex futures, and options.
This paper addresses how asymmetric information, mispricing,
nd jumps in both the price of a risky asset and its mispricing affect
he optimal portfolio strategies and maximum expected utilities of
wo distinct classes of rational long-horizon investors in an economy
here preference is logarithmic. For the purpose of this exposition,
e take the risky asset to be stock, but the model can be applied to any
sset (e.g. currencies) where prices are mean-reverting. The investors
ssume that the risky asset has a fundamental or true value (for ex-
mple, expected discounted future dividends) as well as a market
rice. When prices move away from its fundamental value, they al-
ays revert to it (see, e.g., LeRoy & Porter, 1981; Poterba & Summers,
988; Shiller, 1981; Summers, 1986). What do we mean by funda-
ental value? Allen, Morris, and Postlewaite (1993) posit that the
undamental value of an asset is the present value of the stream of
he market value of dividends or services generated by this asset. The
undamental value can also be defined as the market price that would
revail if the cost of gathering and processing information is zero for
weeney (2006), Serban (2010); indices: Miller, Muthuswamy, and Whaley (1994),
alvers, Wu, and Gilliland (2000), Balvers and Wu (2006), Caporale and Gil-Alana
2008), Kim, Stern, and Stern (2009), Chen and Kim (2011), Spierdijk, Bikker, and
an den Hoek (2012), Malin and Bornholt (2013); index futures: Monoyios and Sarno
2002), Miao, Lin, and Chao (2014).
W. S. Buckley, H. Long / European Journal of Operational Research 243 (2015) 944–955 945
a
a
m
i
p
p
p
o
j
B
a
p
a
o
a
t
b
f
a
t
t
f
i
l
c
j
a
s
s
t
I
e
i
t
u
b
v
T
f
t
o
G
f
o
b
v
o
p
a
t
(
t
p
m
N
T
P
F
T
a
t
o
e
t
i
h
f
h
t
i
m
t
w
s
t
w
l
i
n
k
H
t
o
I
o
d
t
a
r
i
a
m
i
e
b
i
2
b
a
p
T
r
T
μT
i
(m
i
i
ll investors. Any of the definitions above will suffice. However, we
dopt the caveat in Allen et al. (1993) and take fundamental value to
ean the value of an asset in normal use, as opposed to some value
t may have as a speculative instrument.
Because mispricing is the difference between the asset’s market
rice and its fundamental value, if the fundamental value or asset
rice jumps, then it follows that the mispricing will jump as well,
rovided the jump components are not identical and cancel each
ther. This is particularly evident in the case of independently driven
ump processes (see, e.g., Applebaum, 2004). All recent studies (e.g.,
uckley et al., 2012; Buckley, Long, & Perera, 2014; Guasoni, 2006)
ssume that mispricing is a continuous O–U process. However, in this
aper, mispricing is no longer purely continuous. Instead, it jumps
nd is driven by a mean-reverting O–U process which has a continu-
us component as well as a discontinuous component generated by
pure-jump Lévy process. As in Buckley et al. (2014), the price of
he risky asset is still subject to Levy jumps. We solve this model for
oth investors and present explicit formulas for their optimal port-
olios and maximum expected logarithmic utilities under reasonable
ssumptions.
Portfolio allocation problems have been extensively studied since
he seminal work of Markowitz (1952). Merton’s (1971) model is
he benchmark of optimal asset allocation in the continuous-time
ramework. There is a rich plethora of portfolio optimization papers2
n various settings. We refer to Buckley et al. (2012, 2014) for related
iterature review and discussion on recent progress on this topic.
In this paper, we find that the optimal portfolio of each investor
ontains excess risky asset that depends on the Lévy measures of both
ump processes (asset price and mispricing), the diffusive volatility,
nd the level of mispricing, as represented by the mean-reversion
peed or time. Under quadratic approximation of the portfolios, we
how that excess asset holdings are dependent on the first two instan-
aneous centralized moments of return (see, e.g., Cvitanic et al., 2008).
n addition, the maximum expected utility from terminal wealth for
ach investor is increased by the presence of jumps in the mispric-
ng. This suggests that investors are better off when mispricing jumps
han when it changes continuously.
Further, we show that the uninformed investor obtains excess
tility from jumps in mispricing. However, as proven in prior studies
y Guasoni (2006) and Buckley et al. (2012, 2014), the informed in-
estor still has positive excess utility over the uninformed investor.
his implies that the informed investor gets more utility from the dif-
usive component driving the asset price process. We also show that
he asymptotic excess optimal expected utility from terminal wealth
f the informed investor has a similar structure to that presented in
uasoni (2006) and Buckley et al. (2012, 2014), but is increased by a
actor directly attributed to the jumps in the mispricing. As in the case
f jumps in stock price only, the mean-reversion rate λ is replaced
y a smaller adjusted mean-reversion rate λ, which depends on the
olatility of the asset price only.
There is also an equivalent representation of excess utility by way
f the mean-reversion speed λ of an adjusted continuous O–U mis-
ricing process which also depends on the volatility of the long-run
sset price. In this framework, the excess optimal expected utility of
he informed investor is greater (less) than its continuous Merton
1971) geometric Brownian motion counterpart if the product of
he diffusive variance and the quadratic variation of the mispricing
rocess is greater (less) than the second instantaneous centralized
oment of return of the jump component of the asset price process.
otwithstanding the presence of asymmetric information, mispricing
2 See Mansini and Speranza (1999), Kellerer, Mansini, and Speranza (2000), Soyer and
anyeri (2006), Celikyurt and Ozekici (2007), Corazza and Favaretto (2007), Cvitanic,
olimenis, and Zapatero (2008), Lin and Liu (2008), Canakoglu and Ozekici (2010),
u, Lari-Lavassani, and Li (2010), Huang, Zhu, Fabozzi, and Fukushima (2010), Yu,
akahashi, Inoue, and Wang (2010), and Buckley et al. (2012, 2014).
(
I
r
m
l
nd jumps, our results show that it still pays to be more informed in
he long-run, unless there is too little or too much mispricing. More-
ver, our results nest those contained in Guasoni (2006) and Buckley
t al. (2014).
The practical, financial, economic and operational implication of
his paper is that when asymmetric information and jumps exist
n both asset price and mispricing, informed and uninformed long-
orizon investors will maximize their expected logarithmic utilities
rom terminal wealth by holding portfolios that contain excess asset
oldings which depend not only on the level of information asymme-
ry and preference, but also on the nature and frequency of the jumps
n both asset price and mispricing as dictated by the governing Lévy
easures. To the best of our knowledge, this paper is the first to study
he utility and portfolio implications for the risky asset by investors
hen asymmetric information, logarithmic preferences, jumps in as-
et price and mispricing are intertwined, and therefore contributes to
he finance and operational research literature.
Our model is related to heterogeneous agents models (HAM),
here investors observe the same information but have different be-
iefs. However, there are two important distinctions. First, switching
s not allowed between investor classes, i.e., an uninformed investor is
ot allowed to become informed, and vice versa. Second, one investor
nows more than the other (see, e.g., Chiarella, Dieci, & He, 2009 and
e, 2012 for a review of the HAM literature).
As in Buckley et al. (2014), our model is different from insider
rading models3 which use enlargement of filtrations to obtain the
ptimal portfolio and utility of the insider trader/informed investor.
n contrast to insider trading models, we specify the price dynamics
f the informed investor in the larger filtration, and then obtain the
ynamics for the uninformed investor by contracting (or restricting)
he larger filtration. We then compute and compare optimal portfolios
nd expected logarithmic utilities for each investor relative to their
espective filtrations.
The remainder of the paper is organized as follows. The model
s introduced in Section 2, including information flows/filtration and
sset price dynamics of investors. In Section 3, expected utilities are
aximized, and optimal portfolios are computed and estimated us-
ng instantaneous centralized moments of returns. Asymptotic and
xcess utilities are presented in Section 4. We conclude in Section 5
y giving directions of possible future research. All proofs are given
n Appendix A.
. The discontinuous mispricing model
The economy consists of two assets—a risk-free asset B, called
ank account or money market (or U.S. Treasury bill), and a risky
sset S, called stock. The risk-free asset earns a continuously com-
ounded risk-free interest rate rt , and has price Bt = exp(∫ t
0 rs ds)
.
he continuous component of the stock’s percentage appreciation
ate is μt , at time t ∈ [0, T], where T > 0 is the investment horizon.
he stock is subject to volatility σt > 0. The market parameters are
t, rt, σt, t ∈ [0, T], and are assumed to be deterministic functions.
he stock’s Sharpe ratio or market price of risk θt = μt−rtσt
is square
ntegrable. The risky asset has price S, and lives on a probability space
Ω, F , P) on which is defined two independent standard Brownian
otions W = (Wt)t≥0 and B = (Bt)t≥0. The stock is viewed by investors
n disjoint classes populated by uninformed and informed investors,
ndexed by i = 0 and i = 1, respectively. Investors have filtrations Kit
3 See, e.g., Pikovsky and Karatzas (1996), Amendinger, Imkeller, and Schweizer
1998), Amendinger, Becherer, and Schweizer (2003), Ankirchner, Dereich, and
mkeller (2006), Ankirchner and Imkeller (2007), who solve this problem for loga-
ithmic and power utilities. Grorud (2000) studies insider trading in a discontinuous
arket, while Goll and Kallsen (2003) give a complete explicit characterization of
og-optimal portfolios without constraints.
946 W. S. Buckley, H. Long / European Journal of Operational Research 243 (2015) 944–955
∞
Z
C
a
d
w
U
w
U
N
n
U
p
a
I
E
P
V
a
p
P
V
a
I
Y
i
t
d
d
w
υ
2
a
a
u
L
a
with K0t ⊂ K1
t ⊂ F , t ∈ [0, T]. All random objects are defined on a
filtered probability space (Ω,F ,Ki, P).Based on the classical Black–Scholes geometric Brownian motion
model (cf. Black & Scholes, 1973) and Merton jump-diffusion model
(cf. Merton, 1971, 1976) as well as some recent models discussed in
Kou (2002), Guasoni (2006) and Buckley et al. (2014), we propose
a new and more general model for the risky asset by incorporat-
ing mean-reversion, mispricing, asymmetric information, and Lévy
jumps. This new model will be more flexible and help capture the
asset price movements including sudden jumps in both the funda-
mental value (through the mispricing process) and market price. It
will also greatly enhance our understanding of the link between as-
set mispricing and asymmetric information in financial markets. We
therefore extend the theory on mispricing models under asymmetric
information in Lévy markets to include the situation where both stock
price and mispricing jump.
We model the jumps in the mispricing by a pure jump Lévy process
Z = (Zt)t≥0 with E(Zt) = 0 and E(Z2t ) < ∞, for all t ∈ [0, T]. Jumps in
the stock price are driven by an independent pure-jump Lévy process
X = (Xt)t≥0. Both jump processes are governed by Poisson random
measures NS and NU defined on R+ × B(R − {0}) that are linked to the
stock (S) and mispricing (U), respectively. These random measures
count the jumps of the stock price S (equivalently, X) and mispricing U
(equivalently, Z), respectively, in the time interval (0, t). The respec-
tive Lévy measures are vS(dx) = ENS(1, dx) and vU(dz) = ENU(1, dz).
We use NU(dt, dz)�= NU(dt, dz)− vU(dz)dt to denote the compensated
Poisson measure.
2.1. The model
The risky asset or stock has price S, with log-return dynamics
d(log St) =(μt − 1
2σ 2
t
)dt + σt dYt + dXt, t ∈ [0, T] (1)
Yt = p Wt + q Ut, p2 + q2 = 1, p ≥ 0, q ≥ 0, (2)
dUt = −λ Ut dt + dLt, λ > 0, U0 = 0, (3)
Lt = Bt + Zt, (4)
Xt =∫ t
0
∫R
x NS(dt, dx), (5)
Zt =∫ t
0
∫R
z NU(dt, dz), E(Zt) = 0, E(Z2
t
)< ∞. (6)
W and B are independent standard Brownian motions. U = (Ut) is a
mean-reverting Ornstein–Uhlenbeck process with mean-reversion
rate λ. Because of the exponential decay associated with mean-
reverting processes, its half-life is H = ln 2λ
. Half-life is a measure of
the slowness of a mean-reverting process, and is measured in trad-
ing days. It may be defined as the average time it takes the mean-
reverting process to get pulled half-way back to its long-term mean.
In this context, half-life is a very important number—it gives an es-
timate of how long we should expect the mispricing to remain far
from zero. Thus, a half-life of 10 days means that it takes 20 trad-
ing days on average for the mispricing to revert to zero. Noting that
half-life is inversely proportion to the mean-reversion rate, we sim-
ply use δ = 1λ
as the measure of mispricing. Thus, half-life H, λ and
δ are equivalent measures of mispricing. Large values of the mean-
reversion time δ (equivalently, small values of λ) correspond to high
levels of mispricing—that is, the mispricing process U takes a long
time to revert to zero, and as such, indicates that mispricing is persis-
tent. The jumps in the mispricing Ut are controlled by the pure-jump
Lévy process Zt , although it also has a continuous component Bt .
Standing assumption. We assume that NS and NU are independent
random measures if NS �= NU . That is, X and Z are independent pure-
jump Lévy processes with E(Zt) = 0, for all t ∈ [0, T] and∫
zvU(dz) <
R. From (6),
t =∫ t
0
∫R
zNU(ds, dz) =∫ t
0
∫R
zNU(ds, dz)− t
∫R
zvU(dz).
learly Z = (Zt) is a martingale provided∫
R zvU(dz) < ∞. From Eqs. (3)
nd (4), we get the dynamics
Ut = −λ Utdt + dBt + dZt, (7)
hich admits the unique solution
t =�= UBt + UZ
t , (8)
here
Bt =
∫ t
0
e−λ(t−s)dBs and UZt =
∫ t
0
e−λ(t−s)dZs. (9)
ote, in this case, that the mispricing process U is a linear combi-
ation of two independent O–U processes: a continuous componentB driven by the Brownian motion B, which is identical to the mis-
ricing process used by Buckley et al. (2014) and Guasoni (2006),
nd a jump component UZ driven by the pure-jump martingale Z.
t is easy to show that for all λ > 0 and t ∈ [0, T], E(UBt ) = 0 and
(UBt )
2 = 1−e−2λ t
2λ. We have a similar result for the process UZ .
roposition 1. Let λ > 0. For each t ∈ [0, T], E(UZt ) = 0,
ar(UZt ) = 1 − e−2λ t
2λ
∫R
z2 vU(dz),
nd limt→∞
Var(UZt ) = 1
2λ
∫R
z2 vU(dz).
We have a similar result for the mean-reverting mispricing
rocess U.
roposition 2. Let λ > 0. For each t ∈ [0, T], E(Ut) = 0,
ar(Ut) =(
1 − e−2λ t
2λ
) (1 +
∫R
z2 vU(dz)
),
nd limt→∞
Var(Ut) = 1
2λ
(1 +
∫R
z2 vU(dz)
).
t now follows from Eqs. (2) and (8) that
t = p Wt + q Ut = p Wt + q
∫ t
0
e−λ(t−s)dBs + q
∫ t
0
e−λ(t−s)dZs,
s a stochastic process consisting of three independent components:
wo are continuous, while one is a pure-jump Lévy process. Taking
ifferentials and importing Eq. (7) yield
Yt = p dWt + q dUt = p dWt + q(−λUtdt + dBt + dZt)
= p dWt + q dBt − λqUtdt + q dZt
= p dWt + q dBt − λqUBt dt − λUZ
t dt + q dZt
= dB1t + υ1,B
t dt + υZt dt + q dZt = dB1
t + υ1t dt + q dZt, (10)
here
B1t
�= p Wt + q Bt, (11)
1t
�= υ1,Bt + υZ
t , υ1,Bt
�= −λ q UBt , υZ
t
�= −λ q UZt . (12)
.2. Filtration of the investors
We expand the filtration developed in Buckley et al. (2012, 2014)
nd Guasoni (2006) to include the information generated by X = (Xt)nd the martingale Z = (Zt). We assume that all filtrations obey the
sual hypothesis. Namely, they are right-continuous and complete.
et F it (i = 1, 0) be the information flow/filtration for the informed
nd uninformed investors in the models where there are no jumps in
W. S. Buckley, H. Long / European Journal of Operational Research 243 (2015) 944–955 947
m
T
F
T
(
K
n
h
K
w
t
U
D
Y
W
L
−
B
w
υ
γ
F
d
w
υ
υ
t
d
s
P
[
o
P
f
2
i
(
d
w
μ
R
d
W
L
a
s
j
T
a
s
c
i
T
v
d
a
w
μR
ispricing or asset price (see Buckley et al., 2012 or Guasoni, 2006).
hese filtrations are defined as
1t
�= σ (Ws, Bs : s ≤ t) = FS, Ut and F0
t
�= σ (Ys : s ≤ t) = FYt .
he filtration of the informed investor in our new framework is K1 =K1
t )t≥0 defined by
1t = F1
t ∨ σ(Zu, Xu : u ≤ t), t ∈ [0, T].
The uninformed investor observes the stock price only and does
ot know what the mispricing is, if it exists. Uninformed investors
ave filtration K0, where K0t ⊂ F is defined by the prescription:
0t = F0
t ∨ σ(Zu, Xu : u ≤ t), t ∈ [0, T],
hich is contained in K1t , the filtration of the informed investor.
We shall develop the dynamics for these investors, starting with
he log-return dynamics for the stock given by Eqs. (1)–(6). Recall that
t = UBt + UZ
t .
efinition 1. Define the continuous part of the process Y by
ct
�= Yt − q UZt = p Wt + q UB
t . (13)
e have the following important result.
emma 1. There exist an K0-Brownian motion B0, and a process υ0, Bt ≡
φ0, Bt , adapted to K0
t , such that for each t ∈ [0, T] and q ∈ [0, 1],
0t = Yc
t +∫ t
0
φ0, Bs ds ≡ Yt − q UZ
t −∫ t
0
υ0, Bs ds, (14)
here λ > 0 is the mean-reversion speed of U, p2 + q2 = 1 and
0, Bt = −λ
∫ t
0
e−λ(t−s)(1 + γ (s))dB0s , (15)
(s) = 1 − p2
1 + p tanh (λps)− 1, p ∈ [0, 1].
rom Eq. (14), we have Yt = B0t + ∫ t
0 υ0, Bs ds + q UZ
t . Therefore,
Yt = dB0t + υ0, B
t dt + q dUZt = dB0
t + υ0, Bt dt − λ q UZ
t dt + qdZt
= dB0t + υ0, B
t dt + υZt dt + qdZt = dB0
t + υ0t dt + qdZt, (16)
here
0t
�= υ0,Bt + υZ
t , υ0,Bt
�= −λ
∫ t
0
e−λ(t−s)(1 + γ (s))dB0s ,
Zt
�= −λ q UZt . (17)
Note that (16) has the same form as (10). We now extract Proposi-
ion 2 in Buckley et al. (2012) concerning υ i, Bt , the number of standard
eviations from the mean return μt , since it will be required in the
equel.
roposition 3. Let λ > 0, i ∈ {0, 1}, p ∈ [0, 1], p2 + q2 = 1 and t ∈0, T].
Let υ0, Bt = −λ
∫ t0 e−λ(t−s)(1 + γ (s))dB0
s and υ1, Bt = −λ q Ut. Then
(0) E[υ i, Bt ] = 0.
(1) Varυ0, Bt = E[υ0, B
t ]2 = λ2∫ t
0 e−2λ(t−s)(1 + γ (s))2ds ≤ 12 λ q2.
(2) Varυ1, Bt = E[υ1, B
t ]2 = λ2 (1 − p2)(1 − e−2λt) ≤ 1
2 λ q2.
(3) Varυ i, Bt = E[υ i, B
t ]2 → λ2 (1 − p)(1 + (−1)i+1p) : as t → ∞.
(4)∫ T
0 E[υ i, Bt ]2dt λ
2 (1 − p)(1 + (−1)i+1p)T as T −→ ∞.
(5) As T −→ ∞, the asymptotic excess cumulative variance of the
vis is∫ T
0
E[υ1, Bt ]2dt −
∫ T
0
E[υ0, Bt ]2dt λ p (1 − p)T ≤ λ
4T.
We have a useful result for the process υ i, which is a generalization
f the process used in Buckley et al. (2012, 2014) and Guasoni (2006).
roposition 4. Let λ > 0, t ∈ [0, T], i ∈ {0, 1} and υ it = υ i,B
t + υZt . Then
or p, q ∈ [0, 1] with p2 + q2 = 1, we have
(1) E[υ it] = 0.
(2) Var[υ1t ] = λ
2 (1 − e−2λ t)(1 + p)(1 + (−1)1p)(1 + ∫R z2 vU(dz)).
(3) Var[υ0t ] = λ2
∫ t0 e−2λ(t−s)(1 + γ (s))2ds +
λ2 q2(
1−e−2λ t
2λ
) ∫R z2 vU(dz).
Moreover,
limt→∞
E[υ it]
2 = λ
2(1 + p)(1 + (−1)ip)
(1 +
∫R
z2 vU(dz)
).
.3. Asset price dynamics for the investors
Let i ∈ {0, 1}. Define μ∗t
�= μt − 12 σ 2
t . The dynamics for the stock
s given by (1) as d(log St) = μ∗t dt + σt dYt + dXt, t ∈ [0, T]. Imposing
2), (3) and (10) yields
(log St) = μ∗t dt + σt
(dBi
t + υ itdt + q dZt
) + dXt
= μ∗t dt + υ i
tσtdt + σtdBit + q σtdZt + dXt
=(μ∗
t − qσt
∫R
xυU(dx)+ υ itσt
)dt + σtdBi
t
+∫
R
q σtzNU(dt, dz)+∫
R
xNS(dt, dx)
= μ∗,it dt + σtdBi
t +∫
R
q σtzNU(dt, dz)+∫
R
xNS(dt, dx),
(18)
here
∗,it = μ∗
t − qσt
∫R
zυU(dz)+ υ itσt
= μt + υ itσt − qσt
∫R
zυU(dz)− 1
2σ 2
t . (19)
emark 1. If NS = NU = N then v = vU = vS, and (18) becomes
(log St) = μ∗,it dt + σt dBi
t +∫
R
x (qσt + 1)N(dt, dx). (20)
e now quote a very useful result on page 47 in Applebaum (2004).
emma 2 (Proposition 1.3.12, Applebaum (2004)). If (N1(t), t ≥ 0)nd (N2(t), t ≥ 0) are two independent Poisson processes defined on the
ame probability space (,F , P) with arrival times (T(j)n , n ∈ N), for
= 1, 2, respectively, then P(T(1)m = T(2)
n for some m, n ∈ N) = 0.
his result tells us that independent Poisson processes do not jump
t the same time. Thus, for random measures NU and NS and Borel
et A ∈ B(R − {0}), NU(t, A) and NS(t, A) are independent Poisson pro-
esses and do not jump together by Lemma 2. We have the following
mportant result.
heorem 1. Let i ∈ {0, 1}. The log-return dynamics (1), for the ith in-
estor, is
(log St) = μ∗, it dt + σtdBi
t +∫
R
qσt z NU(dt, dz)+∫
R
x NS(dt, dx),
(21)
nd its percentage return dynamics is
dSt
St−= μi
tdt + σtdBit +
∫R
(eqσt z − 1)NU(dt, dz)
+∫
R
(ex − 1)NS(dt, dx), (22)
here
∗, it
�= μit − 1
2σ 2
t , μit
�= μt + υ itσt − qσt
∫zυU(dz).
948 W. S. Buckley, H. Long / European Journal of Operational Research 243 (2015) 944–955
3
s
A
πs
c
p
t
E
T
o
i
t
m
a
3
s
a
t
G
w
i
d
S
t∫D
G
L
T
v
n
T
t
t
π
I
π
w
KR
The asset’s percentage return dynamics (22) consists of four com-
ponents: the first is continuous; the second is random and driven
by a standard Brownian motion; the third is driven by jumps in the
mispricing process and the fourth component is driven by jumps in
the asset price. However the total mean return is driven only by three
components: the first is continuous, the second is due to the jumps
in mispricing, and the third is due to jumps in the asset price.
3. Optimal portfolios and expected utilities
Random objects for the ith investor live on a filtered space
(Ω,F ,Ki, P) for i ∈ {0, 1}.
Definition 2 (Portfolio process). The process π ≡ π i : [0, T] −→ R is
called the portfolio process of the ith investor, if π it(ω) is Ki
t-adapted
for each ω ∈ and E∫ T
0 (π itσt)2dt < ∞.
Note that π it is really π i
t(ω), where ω ∈ , and hence, is a random
process. π it is the proportion of the wealth of the ith investor that is
invested in the risky asset at time t ∈ [0, T]. The remainder, 1 − π it , is
invested in the risk-free asset or money market. Where it is clear, we
drop the superscript “i” and simply use π for the portfolio process.
Definition 3 (The wealth processes). The wealth process for the ith
investor is Vi, π , x : [0, T] −→ R, where Vi, π, xt is the value of the port-
folio consisting of the risky asset and money market at time t, when
π it is invested in the risky asset and the initial capital is x > 0.
For brevity, we denote this process by Vi, π = (Vi, πt ) or simply
Vi = (Vit), t ∈ [0, T], when the context is clear. The initial capital is
Vi0 = x > 0.
Definition 4 (Admissible portfolio). π is called an admissible portfo-
lio if Vπt > 0 almost surely, for all t ∈ [0, T].
3.1. Maximization of logarithmic utility from terminal wealth
Let π ≡ π i ≡ (π it), t ∈ [0, T], be the portfolio process of the ith
investor. Let θ i = μi−rσ be the Sharpe ratio. The corresponding wealth
process with initial capital x > 0 is: Vit ≡ Vi,π
t ≡ Vi,π,xt . We now give
the dynamics for Vi.
Theorem 2. The percentage return of the wealth Vi of the ith investor
with stock price dynamics (1) is:
dVit
Vit−
= (rt + πt σt θ
it
)dt + πt σt dBi
t +∫
R
πt(eqσt z − 1)NU(dt, dz)
+∫
R
πt(ex − 1)NS(dt, dx). (23)
The discounted wealth process is
Vit = Vi
t exp
(−
∫ t
0
rsds
)= Vi
0 exp
(∫ t
0
(πsσsθ
is − 1
2π2
s σ 2s
)ds +
∫ t
0
πsσsdBis
)×�0≤u≤t(1 + πu(e
qσu Zu − 1))× �0≤s≤t(1 + πs(e Xs − 1)).
(24)
Remark 2. We use the subscripts “u” and “s” in Eq. (24) to distin-
guish the jumps due to mispricing U and asset price S, respectively.
The budget constraint (23) consists of four components: the first is
continuous; the second is random and drive by a standard Brownian
motion; the third is driven by jumps in the mispricing process and
the fourth component is driven by jumps in the asset price. Note that
volatility plays a key role in the component driven by the mispricing
process, while the component driven by the jumps in asset price is
independent of volatility. The discounted wealth process has identi-
cal dependencies. This will be the argument used to maximize the
expected logarithmic utility function.
.2. Admissible set for each investor
Let Vi0 = x. We seek a portfolio process π i = (πt)i
t≥0 in an admis-
ible set Ai(x), defined by
i(x)={π i: π i isKi − predictable such that V i, π i
t > 0 a.s.∀t ∈ [0, T]}.i is predictable if it is measurable with respect to the predictable
igma–algebra on [0, T] × , which is the sigma–algebra of all left
ontinuous functions with right limits on [0, T] × . The optimal
ortfolio for the informed or uninformed investor is π ∗,i ∈ Ai(x) such
hat(log V i, π ∗,i
T
) = maxπ∈Ai(x)
E log V i, πT .
hat is, π ∗,i = arg maxπ∈Ai(x) E log V i, πT . Since θt ≡ θ i
t is independent
f πt , observe that E log V i, πT is maximized iff E
∫ T0 f (i)
U,S(πit)dt is max-
mized. That is, if f (i)U,S(π
i) defined by (31) below is maximized on
he admissible set Ai(x). This approach is similar to the optimization
ethod used by Amendinger et al. (1998), Ankirchner et al. (2006),
nd Liu, Longstaff, and Pan (2003).
.3. The objective function G
In order to obtain optimal portfolio strategies, we need the as-
istance of an objective function G. For any Lévy measure va(·),∈ {U, S}, define the (partial) objective function Ga by the prescrip-
ion Ga : [0, 1] −→ R,
a(α ; s)�=
∫R
log(1 + α(es x − 1))va(dx), α ∈ [0, 1],
s ∈ Iσ�= (0, σmax] ∪ {1}, (25)
here s is a non-negative parameter or function representing volatil-
ty and σmax is positive. Note that α, and hence π , is restricted to the
omain [0, 1].
tanding assumption. We assume that an integer k ≥ 2 exists such
hat
R
(e±s x − 1
)kva(dx) < ∞, where a ∈ {U, S}, s ∈ Iσ .
efine the objective function G : [0, 1] −→ R by
(α ; s)�= GU(α ; s)+ GS(α ; 1). (26)
emma 3. Let s ∈ Iσ . If α ∈ [0, 1], then G′′(α ; s) < 0.
his result proves that G is strictly concave on [0, 1] for each fixed
olatility level s, and hence, will admit unique optimal portfolios. We
ow give a major result for optimal portfolios strategies of investors.
heorem 3. Let i ∈ {0, 1} and q ∈ [0, 1]. For the ith investor, the op-
imal portfolio π i, that maximizes the expected logarithmic utility from
erminal wealth over the investment period [0, T], is given by
it = μi
t − rt + G′(π it ; qσt
)σ 2
t
≡ μit − rt + G′
U
(π i
t ; qσt
) + G′S
(π i
t ; 1)
σ 2t
.
(27)
n terms of the stock’s total return bit, the optimal portfolio is
it = bi
t − rt − KU(qσt)− KS(1)+ G′(π it ; qσt
)σ 2
t
, (28)
here
a(s) =∫
(es x − 1)va(dx), a ∈ {U, S}, s ≥ 0. (29)
W. S. Buckley, H. Long / European Journal of Operational Research 243 (2015) 944–955 949
T
x
u
w
G
R
R
πl
n
w
3
s
s
s
d
A
I
V
w
V
T
π
w
G
O
p
a
m
d
a
3
c
s
t
t
i
M
w
K
F
n
σ
M
w
K
f
αn
t
G∫p
G
W
L
he maximum expected logarithmic utility from terminal wealth, with
> 0 in initial capital, is
i(x) = log x + 1
2E
∫ T
0
(θ i
t
)2dt + E
∫ T
0
f (i)U,S
(π i
t
)dt, (30)
here
f (i)U,S(π
i) = G(π i ; qσ)− 1
2(π iσ − θ i)2, (31)
(π i ; qσ) = GU(π i ; qσ)+ GS(πi ; 1),
μit = μt + υ i
tσt − qσt
∫R
xυU(dx),
υ it = υ i,B
t + υZt ,
υ1,Bt = −λ q UB
t , υ0,Bt = −λ
∫ t
0
e−λ(t−s)(1 + γ (s))dB0s ,
υZt = −λ q UZ
t ,
bit = μi
t + KU(qσt)+ KS(1). (32)
emark 3.
(a) The optimal portfolio strategy for each investor consists of
three components. First, there is a continuous component
which is the Merton (1971) optimal for that investor. The sec-
ond component is due to jumps in the mispricing process, level
of mispricing q and volatility. The third component is due to
jumps in the asset price only. One may also present the optimal
portfolio strategy of each investor through it total expected
return bt , the derivative of objective function G and kernel,
Ka.
(b) Let uiT,c(x) = log x + 1
2 E∫ T
0 (θ it)
2dt and uiT,d
(x) = E∫ T
0 f (i)U,S(π
it)dt.
Then ui(x) ≡ uiT(x) = ui
T,c(x)+ uiT,d
(x), where uiT,c(x) is the con-
tinuous component representing the maximum expected loga-
rithmic utility from terminal wealth for the purely continuous
Merton case, and uiT,d
(x) is the discontinuous component which
is due to jumps in the mispricing and the asset prices.
emark 4. Because of the intractability of the integrals∫
R log(1 +(eσ x − 1))va(dx), where a ∈ {U, S}, we compute the optimal portfo-
io π i, using approximation techniques. This is done via the instanta-
eous centralized moments of return for both Lévy processes X and Z,
ith Lévy measures vS and vU , respectively.
.4. The case NU = NS = N
We now study the case where X and Z, given by (5) and (6), re-
pectively, have the same Poisson random measure N, and hence the
ame Lévy measure v = vU = vS. As already demonstrated by (20), the
tock has log-return dynamics
(log St) = μ∗,it dt + σt dBi
t +∫
R
x (qσt + 1)N(dt, dx).
pplication of Itô’s formula yields
dSt
St−= μi
t dt + σt dBit +
∫R
(e(qσt+1)x − 1)N(dt, dx).
t follows from Theorem 1, that the dynamics of the wealth processi is
dVit
Vit−
= (rt + πt σt θ
it
)dt + πt σt dBi
t +∫
R
πt(e(qσt+1)x − 1)N(dt, dx),
ith explicit solution
it = Vi
0 exp
(∫ t
0
rs ds +∫ t
0
(πsσsθ
is − 1
2π2
s σ 2s
)ds +
∫ t
0
πsσsdBis
)× �0≤s≤t(1 + πs(e
(qσs+1) Xs − 1)).
ahe optimal portfolio π i is given by the equation
i = σθ i + G′(π i; qσ)
σ 2= μi − r + G′(π i; qσ)
σ 2,
here
(α ; σ) =∫
R
log(1 + α(e(σ+1)x − 1))v(dx).
bserve that the optimal portfolio strategy for each investor now de-
ends on two components: a Merton (1971) type optimal component
nd a jump component that depends on the jump intensity, level of
ispricing, and the volatility of the diffusive component of the price
ynamic. The budget constraint also has a similar dependence, in
ddition to having a diffusive component.
.5. Approximation of optimal portfolios using instantaneous
entralized moments
We now develop the tools required to estimate optimal portfolio
trategies of investors. We adopt the definition of instantaneous cen-
ralized moments of return as presented in Cvitanic et al. (2008). For
he risky asset with price S and positive integer k, we define the kth
nstantaneous centralized moment by the prescription
S(k)�=
∫R
(ex − 1)kvS(dx) =k∑
j=1
(−1)k−j
(k
j
)KS(j),
here the kernel KS is defined by
S(s)�=
∫R
(es x − 1)vS(dx), s ≥ 0.
or the jump component of the mispricing process U, the instanta-
eous centralized moments of return are dependent on the volatility
t , and are given by:
U(k ; σ)�=
∫R
(eσ z − 1)kvU(dz) =k∑
j=1
(−1)k−j
(k
j
)KU(j ; σ),
here the kernels are
U(j ; σ)�=
∫R
(ej σ z − 1)vU(dz).
We use approximation methods to compute the optimal port-
olios since the partial objective functions Ga(α ; s)�= ∫
R log(1 +(es x − 1)va(dx), a ∈ {U, S} , s ∈ Iσ ≡ (0, σmax] ∪ {1}, is in general
ot tractable. We assume, as before, that∫
R(e±s x − 1)2va(dx) < ∞o ensure that the optimal portfolios exist (i.e.; G′′(α) < 0). Since
(α ; s) = GU(α ; s)+ GS(α ; 1), where α ∈ [0, 1] and s ∈ Iσ , then if
R(e± sx − 1)kva(dx) < ∞, we can approximate G(α ; s)by a kth degree
olynomial Gk(α ; s), where
k(α ; s)�=
k∑j=1
(−1)j−1M(j ; s)αj
j!, (33)
M(j ; s)�= MU(j ; s)+ MS(j ; 1), j = 1, 2, · · · , k.,
Ma(j ; s) =∫
R
(es x − 1)jva(dx) =j∑
r=1
(−1)r−1
(j
r
)Ka(r ; s),
Ka(j ; s) =∫
R
(ej s x − 1)va(dx), a ∈ {U, S}.e have the following lemma.
emma 4. Let k ∈ N. If∫
R(e±s x − 1)kva(dx) < ∞, where s ∈ Iσ , and
∈ {U, S}, then the kth instantaneous centralized moment exists and is
950 W. S. Buckley, H. Long / European Journal of Operational Research 243 (2015) 944–955
T
c
l
γ
T
i
i
c
t
j
t
s
t
a
c
m
d
t
T
s
a
w
t
H
h
i
t
t
W
a
T
i
w
given by
M(k ; s) =k∑
j=1
(−1)j−1
(k
j
)K(j ; s) = MU(k ; s)+ MS(k ; 1),
where
K(j ; s) = KU(j s)+ KS(j), Ka(γ ) =∫
R
(eγ x − 1)va(dx), a ∈ {U, S}.Note that the kernel Ka, where a ∈ {U, S}, is used to calculate the in-
stantaneous centralized moments. Because of the intractable nature
of the integrals involved in computing the optimal portfolio strate-
gies, we resort to approximations of these portfolios using the instan-
taneous centralized moments. The following approximation result is
a direct consequence of Theorem 3.
Theorem 4. Let i ∈ {0, 1} and π i be the optimal portfolio that maximizes
the expected logarithmic utility from terminal wealth over the investment
period [0, T]. Then π(k)i, the approximation of the optimal portfolio
based on a kth degree polynomial approximation (33) of the objective
function G, is given by
πt(k)i = μi
t − rt + G′k(πt(k)i ; qσt)
σ 2t
. (34)
The maximum expected logarithmic utility from terminal wealth, ui(x),with x > 0 as initial capital, is approximated by
u(k)i(x) = log x + 1
2E
∫ T
0
(θ it)
2dt + E
∫ T
0
f (i)U,S(πt(k)
i)dt, (35)
where f (i)U,S(π
i) is defined by (31).
Remark 5. Note from Eq. (34) that the excess stock holding overμi
t−rt
σ 2t
,
the Merton (1971) optimal, isG′
k(πt(k)
i ; qσt)
σ 2t
, which is dependent on
M(j ; s) = MU(j ; s)+ MS(j ; 1), j = 1, 2, · · · , k, the instantaneous cen-
tralized moments of return contained in Eq. (33). Each moment is the
sum of the instantaneous centralized moments for the mispricing and
asset price, respectively.
4. Asymptotic utilities
In this section we discuss the asymptotic results of the optimal
expected utilities for both informed and uninformed investors. We
set the interest rate r = 0 without loss of generality.
As before, we assume that the objective function G(α) ≡ G(α ; s) =GU(α ; s)+ GS(α ; 1) is restricted to the domain [0, 1], where s ∈ Iσ ,
and that∫
R(e±s x − 1)2va(dx) < ∞, for each a ∈ {U, S}. That is, we have
at least a quadratic approximation of G(α). Let Ma(j ; s) = ∫R(es x −
1)jva(dx), a ∈ {U, S}. Define
M(j ; σt)�= MU(j ; σt)+ MS(j ; 1), j = 1, 2, · · · , k, (36)
At�= −M(2 ; qσt)
2(σ 2t + M(2 ; qσt))
,
Bt�= 2M(1 ; qσt)
2(σ 2t + M(2 ; qσt))
,
Ct�= M2(1 ; qσt)
2(σ 2t + M(2 ; qσt))
.
Note that A, B, and C are functions of the diffusive volatility σt . Let
Q(θ) be defined by
Q(θ)�= Aθ2 + Bθ + C, (37)
where θ = μσ represents Sharpe ratio. Let
γ 2t = σ 2
t
σ 2t + M(2 ; qσt)
. (38)
hen γt is the proportion of total volatility due to the diffusive
omponent of the risky asset with dynamics (1). We assume that
imt→∞σt = σ∞ = σ > 0, and set
2 = limt→∞
γ 2t = σ 2
∞σ 2∞ + M(2 ; qσ∞)
. (39)
heorem 5. Assume that∫
R(e±s x − 1)kv(dx) < ∞, s ∈ Iσ , k ≥ 2 and G
s restricted to [0, 1]. Let x > 0 be the initial wealth of investors and
∈ {0, 1}.
(a) As T → ∞, the asymptotic optimal expected utility for the ith
investor due to jumps is
uiT, d(x) ∼
∫ T
0
Q
(μt
σt: σt, M(1; qσt)M(2; qσt)
)dt
+ λ
2A∞(1 − p)(1 + (−1)i+1p)
[1 +
∫R
z2 vU(dz)
]T,
(40)
where
A∞ = limt→∞
At = − M(2; qσ∞)
2(σ 2∞ + M(2; qσ∞)
) . (41)
(b) The asymptotic excess optimal expected utility of the uninformed
investor relative to the informed due to jumps is strictly positive
and given by
u0T, d(x)− u1
T, d(x) ∼ −λ A∞ p (1 − p)
[1 +
∫R
z2 vU(dz)
]T.
(42)
Observe that the optimal expected utility is the sum of two disjoint
omponents—one is continuous, while the other is discrete. uiT, d
(x) is
he discrete component of optimal expected utility. It is driven by the
ump processes only. Part (a) shows that the jump component of op-
imal expected utility depends on the diffusive volatility σt , first and
econd instantaneous centralized moments, M(1; qσt) and M(2; qσt),he level of mispricing q, the jump intensity of the mispricing process,
nd the long-run proportion of total volatility due to the diffusive
omponent of the asset price dynamics. Part (b) gives the excess opti-
al expected utility of the uninformed investor due to jumps only. It
epends on the proportion of mispricing, the mean-reversion speed,
he variance of the mispricing process, and a long-run variance ratio.
his excess is positive because A∞, the long-run proportion of diffu-
ive variance to total variance of the percentage return of the risky
sset is always negative, unless there is no diffusive component, in
hich case it is zero. This is an unexpected result, which suggests
hat jumps favour the uninformed investor from a utility standpoint.
owever, as the next major result shows, the informed investor still
as positive excess utility relative to the uninformed investor. This
mplies that the informed investor gains most of his/her utility from
he diffusive component of the asset price dynamics, and that it pays
o be more informed unless there is too little or too much mispricing.
e have the following major result as a consequence of the quadratic
pproximation of G(α ; σ).
heorem 6. Assume that the conditions of Theorem 5 hold. Let the
nvestment horizon T −→ ∞. Under quadratic approximation of G(α ; s),e have
(a) The asymptotic maximum expected logarithmic utility from ter-
minal wealth for the ith investor, with initial capital x > 0, is
uiT(x) ≈ log x + 1
2
∫ T
0
(μt
σt
)2
dt +∫ T
0
Q
(μt
σt
)dt
+ λ
4(1 − p)(1 + (−1)i+1p)
[1 +
∫R
z2 vU(dz)
]T.
W. S. Buckley, H. Long / European Journal of Operational Research 243 (2015) 944–955 951
l
c
l
m
l
i
o
o
a
h
I
e
i
i
(
m
e
a
s
d
j
λt
O
c
o
b
u
p
c
t
i
t
t
m
2
m
o
i
m
a
c
t
c
5
i
w
a
t
r
p
s
i
m
p
o
t
v
i
c
u
v
g
c
t
n
w
(
m
o
r
i
t
l
t
p
t
g
t
c
n
i
u
l
c
p
t
p
o
t
i
i
t
m
e
w
(b) The asymptotic excess expected logarithmic utility of the informed
investor is
u1T(x)− u0
T(x) ≈ λ
2p (1 − p)
[1 +
∫R
z2 vU(dz)
]T,
where λ = λγ 2 is the long-run adjusted mean-reversion speed
given.
(c) Let λ be the speed of the original mispricing process. There exists
an adjusted continuous O–U mispricing process with speed λ =[1+∫
R z2 vU(dz)]σ 2∞σ 2∞+M(2 ; qσ∞)
, and a continuous diffusion process linked to the
original asset price process with volatility σt , such that the excess
optimal expected utility of the informed investor is
u1T(x)− u0
T(x) ≈ λ
2p (1 − p)T.
Moreover, the excess utility is greater (less) than its continuous ge-
ometric Brownian motion (GBM) counterpart if the product of σ 2∞and the quadratic variation
∫R z2 vU(dz) of the jump component
of U is greater (less) than M(2 ; qσ∞), the second instantaneous
centralized moment of return of the asset price jump process in
the long-run.
Observe from Part (a) that under quadratic approximation, the
ong-run optimal expected utility of each investor is the sum of four
omponents. The first two components are continuous, while the
ast two components depend on the jumps in both asset price and
ispricing processes through their respective Lévy measures. The
evel of mispricing q, its adjusted speed λ, diffusive volatility σt , and
nvestment period T are also important ingredients that influence
ptimal expected utility.
Part (b) presents a version of the excess optimal expected utility
f the informed investor over the uninformed that depends on the
djusted mean-reversion speed λ of the mispricing process, which
as been dampened by the total variance of the asset price process.
t also depends on the level of mispricing and the Lévy measure gov-
rning the mispricing process. Our model shows that the informed
nvestor has more long-run utility, and therefore, it pays to be more
nformed in the long-run, unless there is too little (q ≈ 0) or too much
q ≈ 1) mispricing in the market. Excess optimal expected utility is
aximized when the percentage of mispricing reaches 75%.
Part (c) is another representation of the long-run excess optimal
xpected utility that is identical in structure to Buckley et al. (2014)
nd Guasoni (2006), and indeed nests the mean reversion speeds pre-
ented therein. In this case, the model is equivalent to a continuous
iffusion with a continuous O–U mispricing process having a new ad-
usted mean-reversion speed λ, that is a function of the original speed
, the diffusive volatility, and variances which are constructed from
he Lévy measures of both asset price and mispricing jump processes.
bserve that if there are no jumps, the model reduces to the purely
ontinuous model of Guasoni (2006).
In our more complete framework, the excess utility can be more
r less than its continuous GBM counterpart and is entirely governed
y the jump intensities. In particular, the excess optimal expected
tility is greater (less) than its continuous GBM counterpart if the
roduct of σ 2∞ and the quadratic variation∫
R z2 vU(dz) of the jump
omponent of U is greater (less) than M(2 ; qσ∞), the second instan-
aneous centralized moment of return of the asset price jump process
n the long-run. In the case where there are no jumps in mispricing,
he quadratic variation due to mispricing is zero and consequently
he excess utility is always less than the continuous GBM bench-
ark. As in prior studies by Guasoni (2006) and Buckley et al. (2012,
014), we obtain maximum excess optimal expected utility when
ispricing in the market is at 75%. It is worth noting that whether
r not jumps exist, there is no utility advantage for the informed
nvestor if the market is totally free of mispricing or completely
ispriced.
Our model could be applied to options pricing that incorporates
symmetric information and mispricing of the underlying asset. It
ould be further generalized to a multi-market setting with at least
wo correlated risky assets, and their respective mispricings pro-
esses. Preferences for investors could be general CRRA type.
. Conclusion
We extend the theory of mispricing models under asymmetric
nformation and logarithmic preferences to the discontinuous frame-
ork where jumps in both the asset price and mispricing process
re incorporated. We obtain analogous but more general results than
hose presented in Buckley et al. (2014) and Guasoni (2006). We de-
ive explicit formulas for the optimal portfolios and maximum ex-
ected logarithmic utilities for both investors under reasonable as-
umptions. In particular, we show that the optimal portfolio of each
nvestor contains excess stock holdings which depend on the Lévy
easure of the jump processes driving both the stock price and mis-
ricing processes, through their instantaneous centralized moments
f return.
Under quadratic approximation of the optimal portfolios, we find
hat the optimal expected utility from terminal wealth for each in-
estor is increased by the presence of jumps in the mispricing. Thus,
nvestors are better off when mispricing jumps than when it changes
ontinuously. We also show that the uninformed investor has more
tility resulting from jumps in mispricing. However, the informed in-
estor still has positive excess utility over the uninformed, which sug-
ests that the informed investor gets more utility from the diffusive
omponent driving the asset price process. Furthermore, we show
hat the asymptotic excess maximum expected utility from termi-
al wealth of the informed investor is λ2 p (1 − p)
[1 + ∫
R x2 vU(dx)]
T ,
hich is similar to that presented in Buckley et al. (2014) and Guasoni
2006), but increased by a factor directly attributed to the jumps in
ispricing and asset price. As in the case of jumps in asset price
nly, the mean-reversion speed λ is replaced by an adjusted mean-
eversion speed λ, which is a scalar multiple of the original speed used
n the Brownian motion case and dampened by the total volatility of
he stock including the component produced by the jumps. Equiva-
ently, the excess optimal expected utility is λ2 p (1 − p)T , where λ is
he mean-reversion speed of an adjusted continuous O–U mispricing
rocess which depends on the volatility of the long-run asset price. In
his framework, the excess optimal utility of the informed investor is
reater (less) than its continuous Merton (1971) GBM counterpart if
he product of σ 2T and the quadratic variation
∫R z2 vU(dz)of the jump
omponent of U is greater (less) than M(2 ; qσT), the second instanta-
eous centralized moment of return of the asset price jump process
n the long-run.
We also show that the uninformed investor has more expected
tility resulting from jumps in mispricing in the long-run, which is
arger than what obtained in the Brownian motion case. Despite this
ounter-intuitive and surprising result, the informed investor still has
ositive excess utility over the uninformed, which clearly suggests
hat the informed investor gets more utility from the diffusive com-
onent of the asset price process. Thus, notwithstanding the presence
f asymmetric information, mispricing and jumps, our results show
hat it still pays to be more informed in the long-run, unless there
s too little or too much mispricing in the market. In this case, the
nformed investor has no utility advantage.
Our study can be extended as follows. First, it could be applied
o specific asset classes such as currencies or indices. Second, our
odel assumes that volatility is constant or deterministic. It is well-
stablished empirically that volatility is stochastic, and hence, it
ould be natural to extend the model to the stochastic volatility
952 W. S. Buckley, H. Long / European Journal of Operational Research 243 (2015) 944–955
S
q
V
T
V
B
(
(
A
P
i
d
w
a
f
d
T
d
case. Third, it could also be generalized to multiple risky as-
sets under constant or stochastic volatility, where investors have
CRRA preferences. Fourth, the pricing of options under asymmet-
ric information and mispricing has not received much attention in
the literature. This could be another fruitful direction for future
research.
Appendix A. Proofs
A.1. Proof of Proposition 1
Proof. By the martingale property of Z, E(UZt ) = E
∫ t0 e−λ(t−s)dZs = 0.
By Itô-isometry, and the fact that t∫
R z2 vU(dz) = E(Z2t ) < ∞, we have
Var(UZt ) = E(UZ
t )2 = E
(∫ t
0
e−λ(t−s)dZs
)2
= E
∫ t
0
e−2λ(t−s)d[Z, Z]s
=∫ t
0
e−2λ(t−s)
∫R
z2 vU(dz)ds
=∫ t
0
e−2λ(t−s)ds
∫R
z2 vU(dx)
= 1 − e−2λ t
2λ
∫R
z2 vU(dz).
The last result follows directly by letting t → ∞.
A.2. Proof of Proposition 2
Proof. By Eq. (8), Ut�= UB
t + UZt , whence E(Ut) = E(UB
t )+ E(UZt ) = 0.
Since UB and UZ are independent processes, then it follows from
Proposition 1 that
E(U2t ) = Var(Ut) = Var(UB
t )+ Var(UZt )
= 1 − e−2λ t
2λ+
(1 − e−2λ t
2λ
) ∫R
z2 vU(dz)
=(
1 − e−2λ t
2λ
) (1 +
∫R
z2 vU(dz)
).
The last result follows directly by letting t → ∞.
A.3. Proof of Lemma 1
Proof. Since Yct
�= Yt − q UZt = p Wt + q UB
t is a continuous Gaussian
process (being the sum of two such independent processes), then by
Hitsuda (1968) there exists an F0-Brownian motion B0 and a pro-
cess φ0, Bt adapted to F0
t , such that B0t = Yc
t + ∫ t0 φ0, B
s ds, where φ0, Bs =
−υ0, Bs = λ
∫ t0 e−λ(t−s)(1 + γ (s))dB0
s . Since K0t = F0
t ∨ σ(Zu, Xu : u ≤ t)
then F0t ⊂ K0
t , and since B0t and φ0, B
t are F0t -adapted, they are K0
t -
adapted. Thus, B0t = Yt − q UZ
t + ∫ t0 φ0, B
s ds = Yt − q UZt − ∫ t
0 υ0, Bs ds.
A.4. Proof of Proposition 4
Proof. We prove the case for i = 1. From Eq. (12),
E(υ1
t
) = E(υ1,B
t
) + E(υZ
t
) = −λ qE(UB
t
) − λ qE(UZ
t
) = 0.
ince UB and UZ are independent O–U processes, then with2 = 1 − p2, we have
ar(υ1
t
) = Var(υ1,B
t
) + Var(υZ
t
) = λ2q2 Var(UB
t
) + λ2 q2 Var(UZ
t
)= λ2q2
(1 − e−2λ t
2λ
)+ λ2 q2
(1 − e−2λ t
2λ
) ∫R
z2 vU(dz)
= λ
2(1 − e−2λ t)q2
(1 +
∫R
z2 vU(dz)
)= λ
2(1 − e−2λ t)(1 − p)(1 + (−1)1+1p)
(1 +
∫R
z2 vU(dz)
)= λ
2(1 − e−2λ t)(1 + p)(1 + (−1)1p)
(1 +
∫R
z2 vU(dz)
).
he proof for i = 0 follows similarly. Specifically, by Itô isometry,
ar(υ0
t
) = Var(υ0,B
t
) + Var(υZ
t
)= λ2Var
∫ t
0
e−λ(t−s)(1 + γ (s))dB0s + λ2 q2 Var(UZ
t )
= λ2
∫ t
0
e−2λ(t−s)(1 + γ (s))2ds
+λ2 q2
(1 − e−2λ t
2λ
) ∫R
z2 vU(dz)
y Part 3 of Proposition 3, Varυ i,Bt = E[υ i,B
t ]2 → λ2 (1 − p)(1 +
−1)i+1p) as t → ∞ and hence Varυ it = E[υ i
t]2 → λ
2 (1 + p)(1 +−1)ip)(1 + ∫
R z2 vU(dz)).
.5. Proof of Theorem 1
roof. Let i = 1. By Eq. (18), the dynamics for the informed investor
s equivalently given by
(I(t)) = (μ∗,1t dt + σt dB1
t +∫
R
qσtz NU(dt, dz)+∫
R
x NS(dt, dx),
(A.1)
here I(t) = log St ⇐⇒ St = elog St = eI(t). Let f (x) = ex. Then f ∈ C2(R)nd by Itô’s formula (cf. Theorem 4.4.7, Applebaum (2004)), we have
rom (A.1) that
(f (It)) = df
dx(It−)+ 1
2
d2f
dx2(It−)d[Ic, Ic]t
+∫
R
(f (It− + q x σt)− f (It−)
)NU(dt, dx)
+∫
R
(f (It− + x)− f (It−)
)NS(dt, dx)
= eIt−
[μ∗,1
t dt + σt dB1t + 1
2σ 2
t dt
]+
∫R
eIt−(eqσtx − 1)NU(dt, dx)
+∫
R
eIt−(ex − 1)NS(dt, dx)
= eIt−
[μ1
t dt + σtdB1t +
∫R
(eqσtx − 1)NU(dt, dx)
+∫
R
(ex − 1)NS(dt, dx)
].
hus, we get
(elog St) = St−
[μ1
t dt + σtdB1t +
∫R
(eqσt z − 1)NU(dt, dz)
+∫
(ex − 1)NS(dt, dx)
]
RW. S. Buckley, H. Long / European Journal of Operational Research 243 (2015) 944–955 953
w
T
A
P
c
N
P
T
T
D
o
V
w
A
P
1
s
G
A
P
s
t
w
l
T
E
T
E
w
f
T
u
T
c
1
w
h
f
w
f
A
P
K
M
A
P
a
E
hich is equivalent to the percentage return equation:
dSt
St−= μ1
t dt + σt dB1t +
∫R
(eqσt z − 1)NU(dt, dz)
+∫
R
(ex − 1)NS(dt, dx). (A.2)
he proof for the uninformed investor is identical.
.6. Proof of Theorem 2
roof. We provide the proof for the informed investor. The crucial
omponent of this proof is Lemma 2, which states that NU(t, A) and
S(t, A) do not jump at the same time, because they are independent
oisson processes on each bounded Borel set A ∈ B(R − {0}). From
heorem 1, the percentage return of V1 is
dV1t
V1t−
= (1 − πt)rtdt + πtdSt
St
= (1 − πt)rt dt + πt μ1t dt + πt σt dB1
t
+∫
R
πt(eqσtz − 1)NU(dt, dz)+
∫R
πt(ex − 1)NS(dt, dx).
= (rt + πt σt θ
1t
)dt + πt σt dB1
t +∫
R
πt(eqσtz − 1)NU(dt, dz)
+∫
R
πt(ex − 1)NS(dt, dx).
his proves Eq. (23). By applying the result on Doleans–
ade/Stochastic exponentials, and using the fact of the independence
f jumps of NU and NS, Eq. (23) yields the solution
1t = V1
0 exp
(∫ t
0
rsds +∫ t
0
(πsσsθ
1s − 1
2π2
s σ 2s
)ds +
∫ t
0
πsσsdB1s
)× �0≤u≤t(1 + πu(e
qσu Zu − 1))× �0≤s≤t(1 + πs(e Xs − 1)),
ith discounted wealth given by (24).
.7. Proof of Lemma 3
roof. Let a ∈ {U, S} and s ∈ Iσ . By the standing assumption,∫
R(e± sx −)2va(dx) < ∞. Thus GU, GS ∈ C2[0, 1], and for all α ∈ [0, 1] and
∈ Iσ , G′′a(α ; s) = − ∫
R(es x−1)2va(dx)(1+α(es x−1))2 < 0. Thus, G′′(α ; s) = G′′
U(α ; s)+′′S(α ; 1) < 0.
.8. Proof of Theorem 3
roof. We provide the proof for the informed investor (i = 1). It is
imilar for the uninformed investor (i = 0). Since utility is logarithmic,
hen using Poisson integration, the utility from terminal (discounted)
ealth for the informed investor is:
og V1T = log x +
∫ T
0
(πsσsθ
1s − 1
2π2
s σ 2s
)ds +
∫ T
0
πsσsdB1s
+∑
0≤u≤T
log(1 + πu(eqσu Zu − 1))
+∑
0≤s≤T
log(1 + πs(e Xs − 1))
= log x +∫ T
0
(πsσsθ
1s − 1
2π2
s σ 2s
)ds +
∫ T
0
πsσsdB1s
+∫ T
0
∫R
log(1 + πu(eqσuz − 1))NU(du, dz)
+∫ T ∫
log(1 + πs(ex − 1))NS(ds, dx)
0 R
aking expectation yields
log V1T = log x + 1
2E
∫ T
0
(θ1
s
)2ds − 1
2E
∫ T
0
(πsσs − θ1
s
)2ds
+∫ T
0
∫R
log(1 + πu(eqσuz − 1))vU(dz)ds
+∫ T
0
∫R
log(1 + πs(ex − 1))vS(dx)ds
hus
log V1T = log x + 1
2E
∫ T
0
(θ1
t
)2dt + E
∫ T
0
f (1)U,S(πt)dt, (A.3)
here
(1)U,S(πt) = −1
2
(πtσt − θ1
t
)2 +∫
R
log(1 + πt(eqσtz − 1))vU(dz)
+∫
R
log(1 + πt(ex − 1))vS(dx)
= GU(πt; qσt)+ GS(πt; 1)− 1
2
(πtσt − θ1
t
)2
= G(πt; qσt)− 1
2
(πtσt − θ1
t
)2.
he value function is
1(x) = maxπ∈A1(x)
E log V1,πT = log x + 1
2E
∫ T
0
(θ1
t
)2dt
+ maxπ
E
∫ T
0
f (1)U,S(πs)ds
= log x + 1
2E
∫ T
0
(θ1
t
)2dt + E
∫ T
0
maxπ
f (1)U,S(πs)ds. (A.4)
hus, the objective function f (1)U,S(π) given by (31) is strictly con-
ave by Lemma 3 if π ∈ [0, 1],∫
R(e±σ x − 1)2va(dx) < ∞ and∫
R(e± x −)2va(dx) < ∞, where a ∈ {U, S}. Thus, f (1)
U,S(π) has a maximum at π1
here ddπ
(f (1)U,S(π))|π=π1 = 0. Dropping the subscripts “U” and “S”, we
ave
′(π) = G′(π ; qσ)− σ (πσ − θ1)
= G′U(π ; qσ)+ G′
S(π ; 1)− σ(πσ − θ1) = 0,
hence Eq. (27) holds. The maximum expected utility u1(x) follows
rom the value function (A.4) with maxπ f (1)U,S(π) = f (1)
U,S(π1).
.9. Proof of Lemma 4
roof. Obviously, Ka(j ; s)= ∫R(ej s x − 1)va(dx)=Ka(j s). Thus K(j ; s) =
U(j ; s)+ KS(j ; 1) = KU(j s)+ KS(j), whence
(k ; s) =k∑
j=1
(−1)j−1
(k
j
)K(j ; s)
=k∑
j=1
(−1)j−1
(k
j
)(KU(j ; s)+ KS(j ; 1))
=k∑
j=1
(−1)j−1
(k
j
)KU(j ; s)+
k∑j=1
(−1)j−1
(k
j
)KS(j ; 1)
= MU(k ; s)+ MS(k ; 1).
.10. Proof of Theorem 5
roof. Part (a): Assume that∫
R(e±s x − 1)2v(dx) < ∞, where s ∈ Iσnd G is restricted to [0, 1]. By Proposition 4, as t → ∞(υ i
t
)2 −→ λ
2(1 − p)(1 + (−1)i+1p)
[1 +
∫z2 vU(dz)
], i ∈ {0, 1}.
R
954 W. S. Buckley, H. Long / European Journal of Operational Research 243 (2015) 944–955
B
B
B
B
B
C
C
C
C
C
F
G
G
G
G
G
H
H
H
H
K
K
K
K
L
L
L
M
M
M
Similar to the proof of Theorem 3 in Buckley et al. (2014), as T → ∞,
we have
uiT, d(x) ∼ E
∫ T
0
Q
(μt
σt: σt, M(1; qσt), M(2; qσt)
)dt +
∫ T
0
AtE(υ i
t
)2dt
∼ E
∫ T
0
Q
(μt
σt
)dt + T lim
t→∞AtE
(υ i
t
)2
= E
∫ T
0
Q
(μt
σt
)dt + T A∞ lim
t→∞E(υ i
t
)2
= E
∫ T
0
Q
(μt
σt
)dt
+ λ
2A∞(1 − p)(1 + (−1)i+1p)
[1 +
∫R
z2 vU(dz)
]T,
where A∞ = limt→∞ At = limt→∞ − M(2;qσt)
2(σ 2t +M(2;qσt))
= − M(2;qσ∞)
2(σ 2∞+M(2;qσ∞)).
Part (b): From part(a), since λ2 A∞
∫R z2 vU(dz)T is common for both
investors, it follows that the excess optimal utility of the uninformed
investor due to the jumps is
u0T, d(x)− u1
T, d(x) ≈ −λ
2A∞[(1 − p)(1 + p)− (1 − p)(1 − p)]
×[
1 +∫
R
z2 vU(dx)
]T
= −λ A∞ p (1 − p)
[1 +
∫R
z2 vU(dz)
]T,
which is non-negative.
A.11. Proof of Theorem 6
Proof. (a) The total optimal asymptotic utility of the ith investor
is uiT(x) = ui
T, c(x)+ uiT, d
(x), where the continuous component
is
uiT, c(x) ∼ log x + 1
2
∫ T
0
(μt
σt
)2
dt + λ T
4(1 − p)(1 + (−1)i+1p)
×(
1 +∫
R
z2 vU(dz)
),
and the discontinuous component is given by Theorem 5 as
uiT, d(x) ∼
∫ T
0
Q
(μt
σt
)dt + λ
2A∞(1 − p)(1 + (−1)i+1p)
×(
1 +∫
R
z2 vU(dz)
)T,
where Q(·) also depends on σt , M(1; qσt) and M(2; qσt).(b) This follows by taking the difference of the optimal utilities for
both investors.
(c) This follows immediately from (b) by replacing
λ[1 + ∫R z2 vU(dz)] with λ.
References
Admati, A. (1985). A noisy rational expectations equilibrium for multi-assets securitiesmarkets. Econometrica, 53, 629–658.
Allen, F., Morris, S., & Postlewaite, A. (1993). Finite bubbles with short sale constraintsand asymmetric information. Journal of Economic Theory, 61, 206–229.
Amendinger, J., Becherer, D., & Schweizer, M. (2003). A monetary value for initial in-formation in portfolio optimization. Finance and Stochastics, 7, 29–46.
Amendinger, J., Imkeller, P., & Schweizer, M. (1998). Additional logarithmic utility of
an insider. Stochastic Processes and their Application, 75, 263–286.Ankirchner, S., Dereich, S., & Imkeller, P. (2006). The Shannon information of filtra-
tions and the additional logarithmic utility of insiders. Annals of Probability, 34(2),743–778.
Ankirchner, S., & Imkeller, P. (2007). Financial markets with asymmetric information:Information drift, additional utility and entropy. In Proceedings of the 6th Ritsumeikan
internal symposium.
Applebaum, D. (2004). Lévy processes and stochastic calculus. Cambridge UniversityPress.
alvers, R., Wu, Y., & Gilliland, E. (2000). Mean reversion across national stock mar-kets and parametric contrarian investment strategies. The Journal of Finance, 55,
745–772.alvers, R. J., & Wu, Y. (2006). Momentum and mean reversion across national equity
markets. Journal of Empirical Finance, 13(1), 24–48.lack, F., & Scholes, M. (1973). The pricing of options and corporate liabilities. The
Journal of Political Economy, 81, 637–654.uckley, W. S., Brown, G. O., & Marshall, M. (2012). A mispricing model of stocks
under asymmetric information. European Journal of Operational Research, 221(3),
584–592.uckley, W., Long, H., & Perera, S. (2014). A jump model for fads in asset prices
under asymmetric information. European Journal of Operational Research, 236,200–208.
anakoglu, E., & Ozekici, S. (2010). Portfolio selection in stochastic marketswith HARA utility functions. European Journal of Operational Research, 201(2),
520–536.
aporale, G. M., & Gil-Alana, L. A. (2008). Mean reversion in the Nikkei, Standard & Poorand Dow Jones stock market indices. Journal of Money, Investment and Banking, 5,
13–27.elikyurt, U., & Ozekici, S. (2007). Multi-period portfolio optimization models in
stochastic markets using the mean-variance approach. European Journal of Oper-ational Research, 179(1), 186–202.
hen, S., & Kim, H. (2011). Nonlinear mean reversion across national stock mar-
kets: Evidence from emerging asian markets. International Economic Journal, 25(2),239–250.
heung, Y., & Lai, K. S. (1994). Mean reversion in real exchange rates. Economics Letters,46, 251–256.
Chiarella, C., Dieci, R., & He, X. (2009). Heterogeneity, market mechanisms and as-set price dynamics. In T. Hens, & K. R. Schenk-Hoppe (Eds.), Handbook of financial
markets: Dynamics and evolution (pp. 277–344). Elsevier.
Corazza, M., & Favaretto, D. (2007). On the existence of solutions to the quadratic mixed-integer mean-variance portfolio selection problem. European Journal of Operational
Research, 176, 1947–1960.Cvitanic, J., Polimenis, V., & Zapatero, F. (2008). Optimal portfolio allocation with higher
moments. Annals of Finance, 4(1), 1–28.Depenya, J., & Gil-Alana, I. A. (2002). Mean reversion in the spanish stock market prices
using fractionally integrated semi-parametric techniques. International Journal of
Theoretical and Applied Finance, 5(6), 645–657.Easley, D., & O’Hara, M. (2004). Information and the cost of capital. Journal of Finance,
59, 1553–1583.Engel, C., & Hamilton, J. D. (1990). Long swings in the dollar: Are they in the data and
do markets know it? American Economic Review, 80(4), 689–713.u, C., Lari-Lavassani, A., & Li, X. (2010). Dynamic mean-variance portfolio selec-
tion with borrowing constraint. European Journal of Operational Research, 200,
312–319.oll, T., & Kallsen, J. (2003). A complete explicit solution to the log-optimal portfolio
problem. Annals of Applied Probability, 13, 774–799.ropp, J. (2004). Mean reversion of industry stock returns in the U.S., 1926–1998.
Journal of Empirical Finance, 11(4), 537–551.rorud, A. (2000). Asymmetric information in a financial market with jumps. Interna-
tional Journal of Theoretical and Applied Finance, 4, 641–659.rossman, S., & Stiglitz, J. (1980). On the impossibility of informationally efficient
markets. American Economic Review, 70, 393–408.
uasoni, P. (2006). Asymmetric information in fads models. Finance and Stochastics, 10,159–177.
e, X. (2012). Recent developments on heterogeneous beliefs and adaptive behaviour offinancial markets. Working paper. Sydney: University of Technology.
ellwig, M. F. (1980). On the aggregation of information in competitive markets. Journalof Economic Theory, 22, 477–498.
itsuda, M. (1968). Representation of Gaussian processes equivalent to Weiner process.
Osaka Journal of Mathematics, 5, 299–312.uang, D., Zhu, S., Fabozzi, F., & Fukushima, M. (2010). Portfolio selection under distribu-
tional uncertainty: A relative robust CVaR approach. European Journal of OperationalResearch, 203(1), 185–194.
ellerer, H., Mansini, R., & Speranza, M. G. (2000). Selecting portfolio with fixed costsand minimum transaction lots. Annals of Operations Research, 99, 287–304.
elly, B., & Ljungqvist, A. (2012). Testing asymmetric information asset pricing models.
Review of Financial Studies, 28(5), 1366–1413.im, H., Stern, L., & Stern, M. L. (2009). Nonlinear mean reversion in the G7 stock
markets. Applied Financial Economics, 19(5), 347–355.ou, S. (2002). A jump-diffusion model for option pricing. Management Science, 48,
1086–1101.eRoy, S., & Porter, R. (1981). The present value relationship. Econometrica, 49(3),
555–574.
in, C. C., & Liu, Y. T. (2008). Genetic algorithms for the portfolio selection prob-lems with minimum transaction lots. European Journal of Operational Research, 185,
393–404.iu, J., Longstaff, F., & Pan, J. (2003). Dynamic asset allocation and event risk. Journal of
Finance, 58, 231–259.alin, M., & Bornholt, G. (2013). Long-term return reversal: Evidence from international
market indices. Journal of International Financial Markets, Institutions and Money, 25,
1–17.ansini, R., & Speranza, M. G. (1999). Heuristic algorithms for the portfolio selection
problem with minimum transaction lots. European Journal of Operational Research,114, 219–233.
arkowitz, H. (1952). Portfolio selection. Journal of Finance, 7, 77–91.
W. S. Buckley, H. Long / European Journal of Operational Research 243 (2015) 944–955 955
M
M
M
M
M
M
M
P
P
S
S
S
S
S
S
W
Y
erton, R. C. (1971). Optimum consumption and portfolio rules in a continuous-timemodel. Journal of Economic Theory, 3, 373–413.
erton, R. C. (1976). Option pricing when underlying stock returns are discontinuous.Journal of Financial Economics, 3, 125–144.
iao, D. W., Lin, X. C., & Chao, W. (2014). Option pricing under jump-diffusion modelswith mean-reverting bivariate jumps. Operations Research Letters, 42(1), 27–33.
iller, M. H., Muthuswamy, J., & Whaley, R. E. (1994). Mean reversion of standard &poor’s 500 index basis changes: Arbitrage-induced or statistical illusion? The Journal
of Finance, 49(2), 479–513.
onoyios, M., & Sarno, L. (2002). Mean reversion in stock index futures markets: Anonlinear analysis. The Journal of Futures Markets, 22(4), 285–314.
ukherji, S. (2011). Are stock returns still mean-reverting? Review of Financial Eco-nomics, 20, 22–27.
ukherji, S. (2012). Mean reversion of abnormal stock returns. The Journal of WealthManagement, 14(4), 122–129.
ikovsky, I., & Karatzas, I. (1996). Anticipative portfolio optimization. Advances in Ap-
plied Probability, 28, 1095–1122.oterba, J., & Summers, L. (1988). Mean reversion in stock prices: Evidence and impli-
cations. Journal of Financial Economics, 21(1), 27–59.
erban, A. F. (2010). Combining mean reversion and momentum trading strate-gies in foreign exchange markets. Journal of Banking & Finance, 34(11),
2720–2727.hiller, R. (1981). Do stock prices move too much to be justified by subsequent changes
in dividends? American Economic Review, 71, 421–436.oyer, R., & Tanyeri, K. (2006). Bayesian portfolio selection with multi-variate random
variance models. European Journal of Operational Research, 171, 977–990.pierdijk, L., Bikker, J. A., & van den Hoek, P. (2012). Mean reversion in international
stock markets: An empirical analysis of the 20th century. Journal of International
Money and Finance, 31(2), 228–249.ummers, L. (1986). Does the stock market rationally reflect fundamental value? Journal
of Finance, 41, 591–601.weeney, R. J. (2006). Mean reversion in G-10 nominal exchange rates. Journal of Fi-
nancial and Quantitative Analysis, 41(3), 685–708.ang, J. (1993). A model of intertemporal asset prices under asymmetric information.
Review of Economic Studies, 60, 249–282.
u, M., Takahashi, S., Inoue, H., & Wang, S. (2010). Dynamic portfolio optimization withrisk control for absolute deviation model. European Journal of Operational Research,
201(1), 349–364.