Downloaded from https://iranpaper.

ir
Australian & New Zealand Journal of Statistics
https://www.tarjomano.com https://www.tarjomano.com

Aust. N. Z. J. Stat. 57(3), 2015, 301–324 doi: 10.1111/anzs.12120

HEDGING BARRIER OPTIONS IN GARCH

MODELS WITH TRANSACTION COSTS

Shih-Feng Huang* and Chan-Yi Tsai

National University of Kaohsiung

Summary
This study proposes a modified strike-spread method for hedging barrier options in general-
ized autoregressive conditional heteroskedasticity (GARCH) models with transaction costs.
A simulation study was conducted to investigate the hedging performance of the proposed
method in comparison with several well-known static methods for hedging barrier options.
An accurate, easy-to-implement and fast scheme for generating the first passage time under
the GARCH framework which enhances the accuracy and efficiency of the simulation is
also proposed. Simulation results and an empirical study using real data indicate that the
proposed approach has a promising performance for hedging barrier options in GARCH
models when transaction costs are taken into consideration.
Key words: calendar-spread; first passage time; static hedging; strike-spread.

1. Introduction
The barrier option is the most actively traded exotic derivative in the market. One
distinctive feature of the barrier option is that it suddenly becomes valuable (or worthless) at
the first passage time at which the option is knocked in (or knocked out). This discontinuity
makes the hedging of barrier options more challenging to the issuing institutions than hedging
plain vanilla options. For example, the commonly used delta-hedging for standard options is
particularly impractical for barrier options (Dupont, 2001). To remedy this problem, many
static hedging methods or dynamic replication strategies with matching higher order ‘Greeks’
have been proposed for barrier options. In finance, the ‘Greeks’ are quantities measuring
the sensitivity of the values of financial derivatives to a change in underlying parameters.
For instance, the quantity designated ‘delta’ measures the rate of change of a derivative with
respect to changes in the underlying asset’s price. The quantity designated as ‘theta’ measures
the sensitivity of an option to the passage of time. The quantity designated ‘gamma’ is the
second derivative of the option value with respect to the underlying price. The empirical
findings of Engelmann et al. (2006) indicate that static hedges can perform significantly
better than dynamic hedges. In addition, since the transaction costs of a dynamic hedging
strategy increase as the frequency of relocation increases, a static portfolio is more attractive
to traders for hedging barrier options (Carr & Chou 1997; Carr, Ellis & Gupta 1998).
Two typical static methods for hedging barrier options have been discussed extensively
in the literature. The first one is the classical calendar-spread method proposed by Derman,
*Author to whom correspondence should be addressed.
Institute of Statistics, National University of Kaohsiung, 811, Kaohsiung, Taiwan.
e-mail: [email protected]
Acknowledgments. The authors are thankful to the referees and the associate editor for their constructive
suggestions. This research was partially supported by the Ministry of Science and Technology of Taiwan
under grant NSC 101-2118-M-390-002.

© 2015 Australian Statistical Publishing Association Inc. Published by Wiley Publishing Asia Pty Ltd.
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

1467842x, 2015, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/anzs.12120 by University Rostock Universitätsbibliothek, Wiley Online Library on [26/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
302 HEDGING BARRIER OPTIONS IN GARCH MODELS

Ergener & Kani (1995), abbreviated as DEK. The DEK method uses various European options
with the same strike price but different maturities to approximate the payoff of a barrier
option. Recently, Chung, Shih & Tsai (2010) proposed a modified DEK method, denoted
by MDEK, to improve the hedging performance of the DEK method by incorporating the
condition of zero theta on the barrier. The second method for hedging the barrier options
which was introduced by Bowie & Carr (1994) is called the strike-spread approach. In
contrast with the calendar-spread method, the strike-spread portfolio consists of European
options that have the same maturity but have different strike prices. In the following, the
strike-spread method is modified and extended to handle more complex exotic derivatives
and more general asset dynamics. For examples see Carr & Chou (1997), Carr et al. (1998)
and Carr & Lee (2009).
In this study we assume that the underlying asset process follows a generalized
autoregressive conditional heteroskedasticity (GARCH) dynamic. GARCH models are
capable of depicting many important features of financial data such as heteroscedasticity,
volatility clustering, asymmetry and fat tails (Engle 1982; Bollerslev 1986; Tsay 2010).
While considerable attention has been given in the literature to the pricing and hedging
schemes of barrier options under the Black-Scholes (BS) framework (see Engelmann et al.
2006, and the references therein for examples), less attention has been devoted to the hedging
method in GARCH models. Engelmann et al. (2006) conducted a comparison study to inves-
tigate the hedging performance of several different types of static hedges proposed before
2006 for barrier options. Their results indicate that the pure calendar-spread approaches for
hedging barrier options are not feasible and the unified static hedges proposed by Nalholm
& Poulsen (2006), designated as the NP method in this paper, are robust across market
scenarios. However Engelmann et al. (2006) did not consider the influence of transaction
costs on the hedging performance. Including transaction costs might well have led to differ-
ent conclusions from the comparison study. This current study takes transaction costs into
consideration when comparing the hedging performance of different hedging strategies for
barrier options. The strategies considered include a new proposed hedging method, which
is called the modified strike-spread (MSS) method. Simulation results show that the NP
and MDEK methods perform better than the other methods if transaction costs are ignored
in the BS model. However, if transaction costs are taken into consideration, then the MSS
method has the best hedging performance uniformly according to a variety of risk measures
especially when the transaction costs charged are high. In addition we extend the NP, MDEK
and MSS methods from the BS model to the GARCH framework. The resulting findings
are similar to those from the BS model. Furthermore we conduct an empirical study based
on a data set of barrier option prices covering the period 2 January 2004 to 27 September
2005. The results of the investigation indicate that the MSS method has the best hedging
performance in the GARCH framework.
Our contribution is two-fold. Firstly, to the best of our knowledge, ours is the first paper
that investigates the hedging performance of the sophisticated static methods for hedging
barrier options under the GARCH framework. Our numerical findings indicate that the
effect of transaction costs on the hedging performance is significant and cannot be ignored,
especially for those strategies consisting of trading component securities many times during
the lifetime of the option. The second contribution of our paper is that we establish an
accurate, easy-to-implement and fast scheme to remedy the overestimation problem of the
naive method when generating the first passage times for GARCH processes. The proposed

© 2015 Australian Statistical Publishing Association Inc.

Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

1467842x, 2015, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/anzs.12120 by University Rostock Universitätsbibliothek, Wiley Online Library on [26/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
S. F. HUANG AND C. Y. TSAI 303

procedure for generating the first passage times helps us to make fair comparisons among
various hedging strategies.
The remainder of this paper is organized as follows. Section 2 reviews some relevant
literature including Carr & Chou (1997), Nalholm & Poulsen (2006), Chung et al. (2010) and
introduces the proposed MSS method. The extensions of the NP, MDEK and MSS methods
to GARCH models are introduced in Section 3. An algorithm for generating first passage
times under GARCH models is also given in Section 3. Comparison studies of different
hedging methods for barrier options under the BS and GARCH models are presented in
Section 4. Section 5 presents an empirical study that investigates the hedging performance
of the proposed MSS and MDEK methods. Conclusions are presented in Section 6. Tables,
figures and computational details are given in the Appendix.

2. Static hedging methods of barrier options

In this section, we briefly introduce several sophisticated static methods for hedging
barrier options under the BS model,

dSt = ( − q)St dt + St dWt ,

where St denotes the underlying asset price at time t,  is the expected return, q is the
dividend yield,  is the instantaneous volatility and Wt denotes Brownian motion. In what
follows we use the up-and-out call option (UOC), as an example to illustrate the methods.
The payoff of a UOC option with strike price K, barrier price B and maturity T is defined to
be UOCT (B, K) = (ST − K)+ I(max0tT St <B) , where x+ = max(x, 0) and I(·) is an indicator
function. In particular, B is assumed to be greater than K since the UOC option is worthless
otherwise.

2.1. The strike-spread method of Carr & Chou (1997)

Carr & Chou (1997) proposed a strike-spread hedging method, designated here as the
CC method. The CC method incorporates some European-style options to replicate the payoff
f of the barrier option as follows:

Vt(1) = fˆ()e−r(T −t) + 

f  ()[Se−q(T −t) − e−r(T −t) ]
   ∞ (1)
+ 
f (h)Pt (h, T )dh + 
f  (h)Ct (h, T )dh,
0 

where Vt(1) is the value of the hedging portfolio at time t, fˆ and fˆ , respectively, are the
first and second derivatives of fˆ, which is the adjusted payoff of f . The expressions Ct (h, T )
and Pt (h, T ), respectively, denote the European call and put options with strike price h and
maturity date T at time t, and  is a pre-determined constant. In particular the adjusted
payoff fˆ can be obtained by removing the requirement that St < B and computing the limit
of the pricing formula of the barrier option as T approaches zero. Carr & Chou (1997) used
the down-and-in call option as an example to illustrate the derivation of fˆ. In this study
we derive the CC hedge portfolio of a UOC option, which contains at the initial time the
following components:

© 2015 Australian Statistical Publishing Association Inc.

Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

1467842x, 2015, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/anzs.12120 by University Rostock Universitätsbibliothek, Wiley Online Library on [26/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
304 HEDGING BARRIER OPTIONS IN GARCH MODELS

long 1 call at strike K,

 B p−2 2
short K
European calls at strike BK ,
p
short B
(B − K) European calls at strike B, (2)
short 2(B − K) units of binary call options at strike B
 h p−1 2
and short B
(p − 1)( p−2
h
− pK
B2
) European calls at strike h for h ∈ [B, BK ]

where p = 1 − 2(r−q)
2
. In other words, the CC hedge for a UOC option consists of plain
vanilla call options and binary call options at different strike prices but the same maturity.
2
In practice one can choose finite strike prices between B and BK to approximate the last
component given in (2). The details of the derivation of the CC hedge for the UOC option
are given in the Appendix.
In addition, we define the hedging loss of the CC method at maturity T by

D(1) = UOCT (B, K) − VT(1) + erT {V0(1) − UOC0 (B, K)}, (3)

where the difference V0 − UOC0 (B, K) denotes the initial hedging costs of the CC method.

2.2. The unified method of Nalholm & Poulsen (2006)

Nalholm & Poulsen (2006) suggested a hedge method to unify the strike-spread approach
and the calendar-spread method of Derman et al. (1995) so as to form approximate hedges
under the stochastic volatility model. They divided the price, time and volatility spaces by grid
points at which the hedge is constructed to match the payoff of a barrier option by combining
several plain vanilla options with different strike prices and maturities. The weights of each
plain vanilla option in the hedging portfolio are solved for using a backwards-recursive
sequence of linear systems. In the following we briefly illustrate the NP hedging method.
We choose n time points t0 ,…, tn−1 , where 0 = t0 < t1 < · · · < tn−1 < tn = T , and call
these the ‘match points’. We also divide the volatility space by 1 ,…, v , where 0 < 1 <
· · · < v < ∞. Basically, the NP hedge is designed to match the barrier option payoff at each
match point ti for all j , j = 1,…, v, if Sti = B, for i = 0,…, n − 1. The NP hedge procedure
consists of the following two steps. The first step is to use the fact that UOCti (B, K) = 0 if
Sti = B for i = 0,…, n − 1, to construct a strike-spread portfolio with values i,j at time ti
with volatility j , where


n−1
i,j = Cti (K, T ; j ) + `,j Cti (K` , T ; j ). (4)
`=0

In (4) Ct (K, T ; ) denotes the time-t value of a European call option with strike K, maturity
T and volatility , and the weights `,j are solved for by setting


n−1
Cti (K, T ; j ) + `,j Cti (K` , T ; j ) = 0 (5)
`=0

for i = 0,…, n − 1 and j = 1,…, v. In (5) Sti is set to be B and the strike prices K` are set to
satisfy B = K0 < K1 < · · · < Kn−1 for computing the values of Cti (K, T ; j ) and Cti (K` , T ; j ).
For further discussion of the selection of Ki , we refer the reader to Nalholm & Poulsen
(2006).

© 2015 Australian Statistical Publishing Association Inc.

Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

1467842x, 2015, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/anzs.12120 by University Rostock Universitätsbibliothek, Wiley Online Library on [26/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
S. F. HUANG AND C. Y. TSAI 305

The second step is to construct a calendar-spread hedge, which consists of call options
with strike price B and different maturities. Let ti,` , ` = 1,…, v, be time points chosen in
the interval (ti , ti+1 ], i = 0,…, n − 1. The corresponding portfolio weights i,1 ,…, i,v of the
components in the calendar-spread hedge at each match point ti are found by solving the
following linear system at Sti = B in backward order from i = n − 1 to 0,
⎛ C (B, t ;  ) · · · C (B, t ;  ) ⎞ ⎛  ⎞ ⎛  ⎞
ti i,1 1 ti i,v 1 i,1 i,1
⎝ .. ..
.
.. ⎠ ⎝ .. ⎠ = ⎝ .. ⎠. (6)
. . . .
Cti (B, ti,1 ; v ) ··· Cti (B, ti,v ; v ) i,v i,v
The quantities i,j , j = 1,…, v on the right hand side of (6) are the values of the options in
the hedge that have already been included, that is,
−i,j , i = n − 1,
i,j = n−1 v (7)
−i,j − `=i+1 k=1 `,k Cti (B, t`,k ; j ), i < n − 1,

where the t,j are defined in (4).

It follows from (4)–(7) that the initial value V0(2) of the NP method with initial volatility
 is obtained as
(0)


n−1
V0(2) = C0 (K, T ; (0) ) + `,j C0 (K` , T ; (0) )
`=0
(8)

n−1 
v
+ i,k C0 (B, ti,k ; (0) ).
i=0 k=1

In particular, under the BS model, if we skip the first step of the NP method and give v its
‘natural’ value of 1, then the NP method reduces to the classical calendar-spread method
of Derman et al. (1995), that is, the DEK method. On the other hand, if the second step
is ignored, then the first step also provides a pure strike-spread hedging portfolio, which
converges to the adjusted payoff of the CC method as the number of match points n tends
to infinity (Nalholm & Poulsen 2006).
Similarly to (3), the hedging loss of the NP method is equal to

D(2) = UOCT (B, K) − VT(2) + erT {V0(2) − UOC0 (B, K)}. (9)

If the barrier is never reached before maturity, that is, if  > T , then VT(2) = CT (K, T ) =
UOCT (B, K) since ST  max0tT St < B. Thus

D(2) I(>T ) = erT {V0(2) − UOC0 (B, K)}I(>T ) . (10)

On the other hand, if   T then UOCt (B, K) = 0 for t ∈ [, T ] and the NP strategy is to clear
all the unexpired derivatives in the hedging portfolio at time . In this case VT(2) = er(T −) V(2)
and

D(2) I(T ) = −er(T −) V(2) I(T ) + erT {V0(2) − UOC0 (B, K)}, (11)

where for  ∈ (ti−1 , ti ], i ∈ {1,…, n}, V(2) = C (K, T ; () ) + n−1

`=0 `,j C (K` , T ;  ) +
()

`=i−1 k=1 `,k C (B, t`,k ;  ) with  being the volatility at time  and C (B, s;  ) = 0
n−1 v () () ()

for s < .

© 2015 Australian Statistical Publishing Association Inc.

Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

1467842x, 2015, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/anzs.12120 by University Rostock Universitätsbibliothek, Wiley Online Library on [26/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
306 HEDGING BARRIER OPTIONS IN GARCH MODELS

2.3. The modified calendar-spread method of Chung et al. (2010)

Chung et al. (2010) proposed a modification of the well-known DEK method of Derman
et al. (1995) which is called the MDEK method. Their results show that the MDEK method
provides significant improvement over DEK for hedging barrier options in practice. The
MDEK method is illustrated in the following.
Let Vt(3) denote the value at time t of the MDEK portfolio with n match time points
at time t where the match time points are defined in the same manner as in Section 2.
Furthermore let Bint (K, T ) denote the price of a binary option having payoff I{ST >K} with
strike price K and maturity T at time t ∈ [0, T ]. Chung et al. (2010) proposed establishing
the following MDEK hedge at the initial time for hedging a barrier option:

n−1
V0(3) = C0 (K, T ) + wi C0 (B, ti+1 ) + ŵi Bin0 (B, ti+1 ), (12)
i=0

where the weights of wi and ŵi are determined from the value-matching and theta-matching
conditions at time ti in backward order from i = n − 1 to 0.
At time tn−1 , if the stock price equals the barrier price B, then solve for wn−1 and ŵn−1
using the value-matching condition
Ctn−1 (K, T ) + wn−1 Ctn−1 (B, T ) + ŵtn−1 Bintn−1 (B, T ) = 0, (13)
and the theta-matching condition
@Ct (K, T ) @Ct (B, T ) @Bint (B, T ) 
+ wn−1 + ŵn−1  = 0. (14)
@t @t @t t=tn−1

In the foregoing
Ct (K, T ) = St e−q(T −t) (d+Å ) − Ke−r(T −t) N (d2Å ),
Bint (K, T ) = e−r(T −t) (d2Å ),
@Ct (K, T )
= qSt N (d+Å )e−q(T − t) − rKe−r(T −t) (d2Å )
@t
St 
− √ e−q(T −t) (d+Å ),
2 (T )
@Bint (K, T ) d−Å
= e−r(T −t) (r(d2Å ) + (d2Å )).
@t 2(T − t)
2 √
In turn, in these expressions, d±Å = and d2Å = d+Å −  T − t.
log SKt +(r−q± 2 )(T −t)
√
 T −t
Similarly to (13) and (14), at time ti , i < n − 1, the weights wi and ŵi are found by
solving

n−1
Cti (K, T ) + w` Cti (B, t`+1 ) + ŵ` Binti (B, t`+1 ) = 0,
`=i

and
 @C (K, T ) 
n−1
@Ct (B, t`+1 ) @Bint (B, t`+1 ) 
t
+ w` + ŵ`  = 0.
@t `=i
@t @t t=ti

© 2015 Australian Statistical Publishing Association Inc.

Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

1467842x, 2015, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/anzs.12120 by University Rostock Universitätsbibliothek, Wiley Online Library on [26/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
S. F. HUANG AND C. Y. TSAI 307

In view of (12), the initial hedging cost of the MEDK method is not zero unless n
tends to infinity. Thus the hedging loss of the MDEK method for a UOC option at maturity
T is given by

D(3) = UOCT (B, K) − VT(3) + erT {V0(3) − UOC0 (B, K)}. (15)

By means of a development similar to that in (9)–(11), we find that

 (3)
D I(>T ) = erT {V0(3) − UOC0 (B, K)}I(>T ) ,
D(3) I(T ) = −er(T −) V(3) I(T ) + erT {V0(3) − UOC0 (B, K)},
where for  ∈ (ti−1 , ti ], i ∈ {1,…, n},

n−1
V(3) = C (K, T ) + w` C (B, t`+1 ) + ŵ` Bin (B, t`+1 ).
`=i−1

2.4. The modified strike-spread method

From (8) and (12) we see that the initial hedge portfolios of the NP and NDEK methods
contain 2n + 1 components under the BS model, where n denotes the number of match time
points. Numerical studies presented by Nalholm & Poulsen (2006) and Chung et al. (2010)
have shown that the NP and MDEK methods are capable of hedging barrier options well
with zero transaction costs especially when n is large. However, if we take transaction costs
into consideration, a large value of n leads to an increase in transaction costs especially
when the investor closes out the hedge portfolio at the first passage time . To remedy this
problem, we propose the following MSS method which uses only three components to form
the hedge portfolio.
Under the BS model, if r = q = 0, then the following identity holds:
K B2
UOC0 (B, K) = C0 (K, T ) − C0 ( , T ) − (B − K)UID0 (B), (16)
B K
where UID0 (B) denotes the price of an up-and-in digital option having payoff I(max0tT St >B)
with barrier B and maturity T . By modifying (16) to accommodate the general case in which
r > q  0, we obtain the following MSS method with initial hedging value
K B2
V0(4) = C0 (K, T ) − C0 ( , T ) − UID0 (B), (17)
B K
where
2
C0 (K, T ) − KB C0 ( BK , T ) − UOC0 (B, K)
= . (18)
UID0 (B)
The proposed method is described as follows: Similarly to the NP and MDEK methods, if
the first passage time   T , then clear the hedging portfolio at time  and invest the profit
in (or withdraw the loss from) a money market account until maturity T . By the definition
of given in (18), we have V0(4) = UOC0 (B, K) and the hedging loss of the MSS method
at maturity T given by

D(4) = UOCT (B, K) − VT(4) . (19)

© 2015 Australian Statistical Publishing Association Inc.

Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

1467842x, 2015, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/anzs.12120 by University Rostock Universitätsbibliothek, Wiley Online Library on [26/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
308 HEDGING BARRIER OPTIONS IN GARCH MODELS

By means of a development similar to that in (9)–(11), we have

D(4) I(>T ) = 0, (20)

D(4) I(T ) = −er(T −) V(4) I(T ) , (21)

where V(4) = C (K, T ) − KB C ( BK , T ) − e−r(T −) .

2

Notice that the case where  > T is more demanding here for hedging the short position
of a UOC option than it is in the case where   T since UOCT (B, K)I(>T ) could be positive
and UOCT (B, K)I(T ) is worthless. The MSS hedge provides a perfect hedge when  > T
as shown in (20). The expression given in (21) provides the hedging loss of the MSS for
  T but it is non-trivial to analyse (21) directly since C (·, ·) is a non-linear function of .
We demonstrate the numerical performance of (21) in Section 4.

3. The first passage time in GARCH models

GARCH models have been shown to outperform the BS model in financial derivative
pricing (Engle & Rosenberg 1994, 1995; Duan 1995; Heston & Nandi 2000). This study
employs the following GARCH(1,1) model to investigate the hedging performance of the
NP, MDEK and MSS methods:
Rt = (t ) − 0.5t2 + t "t ,
(22)
t2 = 0 + 1 t−1 "t−1 + 2 t−1 ,
2 2 2

where Rt = log(St =St−1 ) denotes the log returns of the stock price process St , (t ) is an
Ft−1 -measurable function with Ft being the information that is available up to time t, "t
are independent and identically distributed (i.i.d.) N (0, 1) random variables, 0 , 1 and 2
are nonnegative and 1 + 2 < 1. Model (22) has been widely discussed in the literature on
financial derivative pricing and hedging. For example, Duan (1995) and Duan & Simonato
(2001) considered the case in which t (t ) = r + t , where r denotes the constant risk
free interest rate and denotes the constant unit risk premium. Heston & Nandi (2000)
considered the case in which (t ) = r + t2 + 0.5t2 .
There is no closed-form formula for pricing barrier options in GARCH models. Thus,
the hedging method proposed by Carr & Chou (1997) can’t be extended to the GARCH
model directly since it needs to use the closed-form representation to derive the adjusted
payoff. On the other hand, the NP, MDEK and MSS methods can be extended to the GARCH
framework by incorporating numerical algorithms to compute the prices of the derivatives
in the hedging portfolios. For example, the empirical martingale simulation of Duan &
Simonato (1998), the dynamic semiparametric approach of Huang & Guo (2009) and the
empirical P-martingale simulation of Huang (2014) and Huang & Tu (2014) can accurately
compute the derivative prices in the GARCH framework.
In order to investigate the performance of the hedging methods, one has to generate
the first passage time  in a GARCH model. The naive method for generating  is to first
generate the stock prices from model (22) and then determine the time when the stock price
first exceeds the barrier. This naive method is simple and straightforward. However, model
(22) only tells us one stock price, say the closing price, on a single day. In reality a UOC
option could already be knocked out at some time during the given day after which time the
price could have gone down so that the closing price is less than the barrier. Consequently

© 2015 Australian Statistical Publishing Association Inc.

Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

1467842x, 2015, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/anzs.12120 by University Rostock Universitätsbibliothek, Wiley Online Library on [26/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
S. F. HUANG AND C. Y. TSAI 309

the naive method will overestimate  and thereby lead to unfair comparison results in our
investigation. To remedy this defect, we propose the following scheme to generate  by
means of incorporating the idea of continuous-time GARCH models. Following Kallsen &
Taqqu (1998), we define the continuous-time GARCH price process St by
 t 
St = S0 exp [(s− ) − 0.5s− ]ds + Ht ,
2
(23)
0
t
where Ht = 0 s− dWs , Wt is standard Brownian motion,  : R+ → R is a given function
with continuous first derivative and with lim supx→∞ (x)=x < ∞, and
0 , for 0  t < 1,
t =
{ 0+ 1 (H[t] − H[t]−1 )
2
+ 2 [t]−1 }
2 1=2
, for t  1.

In the foregoing [t] denotes the largest integer less than or equal to t. In order to see that
model (23) is in fact an extension of (22), observe that for integer values of t, log(St =St−1 ) =
(t− ) − 0.5t−
2
+ t− (Wt − Wt− ) and

02 , for t = 1,
t−
2
=
0+ 1 (t−1)− (Wt−1 − Wt−2 ) + 2 (t−1)− , for t  2.
2 2 2

Thus, for n < t < n + 1, the price process St behaves like geometric Brownian motion with
drift (n ) and volatility n . The parameter n is in fact random and changes from one
integer time to the next according to the second identity in (22). This continuous-time model
motivates us to propose the following scheme for generating the first passage time under
GARCH models.
Let Xt = log St and let dXt = ( − 0.52 )dt + dWt . According to results from Atiya
& Metwally (2005), one can compute the conditional probability of supt−1st Ss  B given
the values of the terminal prices St−1 and St . That is

Pt = Pr(sup Xs  D|Xt−1 = log St−1 , Xt = log St )

t−1st
(24)
exp{− 2(Xt−1 −D)(X t −D)
}, if Xt < D,
= t2
1, otherwise,

where D = log B and t = 1=T . The proposed procedure for simulating  for GARCH
models is as follows:
(j)
1. Generate m paths of stock prices from (22) and mark them by St , for j = 1,…, m and
t = 1,…, T .
2. For each time interval [t − 1, t] of the jth path, let

dXs(j) = ((s−1)− ) − 0.5(s−1)−

2
+ (s−1)− (Ws − W(s−1)− ),

for s ∈ [t − 1, t]. By (24), one obtains the conditional probability,

 
(j) (j) (j) (j) (j)
Pt = Pr sup Xs(j)  D|Xt−1 = log St−1 , Xt = log St ,
t−1st

for the jth path.

© 2015 Australian Statistical Publishing Association Inc.

Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

1467842x, 2015, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/anzs.12120 by University Rostock Universitätsbibliothek, Wiley Online Library on [26/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
310 HEDGING BARRIER OPTIONS IN GARCH MODELS

(j) (j)
3. Starting with t = 1 generate a Bernoulli random variable t with probability Pr( t =
(j) (j)
1) = Pt . If t = 1, then conclude that ˆ ∈ [t − 1, t] and stop. Otherwise go to
step 4.
(j)
4. Set t = t + 1 and repeat step 3 until t = 1 or t = T . If ˆ ∈ [0, T ], set ˆ = ∞.

Note that Step 1 in the above procedure is also required in the naive method for
generating the first passage time. Steps 2–4 are used to remedy the overestimation problem
in the naive method. The computational burden of Steps 2–4 is tiny since (24) provides
the closed-form representation of the conditional probability of {supt−1st Ss  B} given
the values of the terminal prices St−1 and St . Thus the proposed procedure is accurate,
easy-to-implement and fast.

4. Comparison study
In this section the hedging performances of the methods introduced in Section 2 are
compared under the BS and GARCH models with (K=S0 , B=S0 ) = (1.0, 1.2) and (1.1, 1.3), and
maturity T = 1 (year). Since transaction costs are taken into consideration in our simulation
we define the adjusted hedging losses with transaction costs of the CC, NP, MDEK and
MSS methods at time T by

HET(i) (k) = D(i) + erT H0(i) (k) + er(T −) H(i) (k), i = 1, 2, 3, 4, (25)

where D(i) , i = 1, 2, 3, 4, respectively, are defined in (3), (9), (15) and (19), H0(i) (k) is the
initial transaction cost of the construction of the ith hedge portfolio and H(i) (k) is the
transaction cost when clearing the ith portfolio at the first passage time . In (25) k is
a percentage that determines the transaction costs for buying or selling securities during
[0, T ]. More specifically, by (17) and using the MSS method as an example for illus-
2
tration, we have H0(4) (k) = k × {C0 (K, T ) + KB C0 ( BK , T ) + | |UID0 (B)} and H(4) (k) = k ×
2
{C (K, T ) + KB C ( BK , T )}. In practice, k is usually greater than 1.5% and can even be more than
3.5% inasmuch as it includes the commission and bid-ask spread (Hull 2012, pp.
154 and 204).

4.1. Hedging performance under the Black-Scholes model

Note that the hedging portfolio of the CC method defined in (2) is impossible to
establish in reality since it contains an infinite number of call options with various strike
prices. Therefore we approximate the last component in (2) by 4i=1 ( hBi )p−1 (p − 1)( p−2
hi
− pK
B2
),
2
where hi = B + ( BK − B)(i − 1)=3. For the NP method, we adopt the following three types
of settings used in Engelmann et al. (2006) to evaluate the performance:

1. The first type is abbreviated as NP0, which consists of three call options in the strike-
spread part and a call option with strike price K. The strike prices of the strike-spread
part are chosen to be K1 = B, K2 = B × 1.01% and K3 = K2 + (B − K). In particular,
NP0 contains 0 hedge option in the calendar-spread component.
2. The second type, which is abbreviated as NP4, contains all of the hedge options in NP0.
In addition, NP4 contains four more call options with strike price B and equidistant
expiries in the calendar-spread component.

© 2015 Australian Statistical Publishing Association Inc.

Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

1467842x, 2015, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/anzs.12120 by University Rostock Universitätsbibliothek, Wiley Online Library on [26/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
S. F. HUANG AND C. Y. TSAI 311

3. The third type is abbreviated as NP9, and is similar to NP4 but has nine hedge options
in the calendar-spread component.
For the MDEK method, we perform our calculations for the cases n = 6 and 100, denoted
by MDEK6 and MDEK100, respectively, to investigate the tradeoff between increasing the
number of monitoring time points n and the associated transaction costs.
Unlike the procedure for generating  for the GARCH model given in Section 3, where
we only have to decide which time period  belongs to, in the BS framework we now need
to estimate the actual value of . Hence we modify the previously used procedure in the
following way. At Step 1, we generate the paths of stock prices as before. At Step 3, if
(j)
t = 1, then we take
`
1  g(uk )=Pt
(j)
ˆbj = uk
` k=1 1=t

as an estimator of , where uk , k = 1,…, l, are l independent values sampled from a

(j) (j)
Uniform(St−1 , St ) distribution and

D − Xt−1  [X − D − ( −  )(t − )]2 2

t
g( ) =  exp − 2
2y 2 3 (t − ) 2(t − ) 2
(26)
[Xt−1 − D + ( − 2 ) ]2 
2

− .
2 2
√
In (26) y = exp{ − [Xt−1 − Xt + ( − 2 =2)t]2 =(2t2 )}= 22 t. We refer the reader
to Atiya & Metwally (2005) for details of the derivation of (26).
Tables 1 and 2 present the simulation results of the CC, NP0, NP4, NP9, MDEK6,
MDEK100 and MSS methods for UOC options with q = 0,  = 0.15, and (K, B) = (110, 130)
and (100,120), respectively. The adjusted hedging losses with transaction costs of each method
are computed by (25) with m = 104 random paths. The rate k of the transaction costs is set
to be 0%, 1%, 2% and 3%.
In order to simplify the notation, let HEj denote the adjusted hedging loss with transaction
costs given by (25), for the jth generated path. We employ the following six different
measures to assess the hedging performance: (i) Mean = m1 m j=1 HEj ; (ii) mean absolute
deviation around the median, Madev = m1 m j=1 |HE j − M |, where M denotes the median of
HEj , j = 1,…m; (iii) the quadratic hedging error, E(HE 2 ) = 1 m
HE 2 ; (iv) the expected
m j=1 j
+ ) = 1
hedging loss, E(HE +
where A1 = {j : HEj > 0, j = 1,…, m} and m1 is the
m1 j∈A1 HEj ,
 0.05 , which denotes the 0.95th order
number of elements in A1 ; (v) the 5% value-at-risk, VaR
statistic of HEj , j = 1,…m; (vi) the 5% expected shortfall, ES  0.05 = 1
m2 j∈A2 HEj , where
 0.05 , j = 1,…, m} and m2 is the number of elements in A2 . For each of
A2 = {j : HEj > VaR
the six performance measures, the hedging method with the smallest value has the best
performance and is indicated by boldface in Tables 1 and 2.
Tables 1 and 2 demonstrate that if there is no transaction cost then the MDEK100
method has the best performance according to most of the measures. The NP0 and NP4
also perform well. However, if transaction costs are positive, then the MDEK6 and the MSS
methods perform better than the other methods, especially when transaction costs become
large. The results show that the MDEK6 and the MSS methods suffer less from an increase

© 2015 Australian Statistical Publishing Association Inc.

Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

1467842x, 2015, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/anzs.12120 by University Rostock Universitätsbibliothek, Wiley Online Library on [26/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
312 HEDGING BARRIER OPTIONS IN GARCH MODELS

Table 1
The hedging performances of the CC, NP0, NP4, NP9, MDEK6, MDEK100 and MSS methods
for UOC options with S0 = 100, r = 0.05, q = 0,  = 0.15, K = 110, T = 365 (days), B = 130
and m = 104 random paths under the BS model.

CC NP0 NP4 NP9 MDEK6 MDEK100 MSS

Case 1: k = 0% UOC0 (B, K) = 1.4960

Mean 0.1346 −0.0277 −0.0278 −0.0230 −0.0085 −0.0000 −0.0007

MAD 2.4134 0.0808 0.0820 0.0906 0.0220 0.0003 0.1817
E(HE 2 ) 35.8244 0.2467 0.2471 0.2502 0.0058 0.0000 0.2231
E(HE + ) 10.5476 0.0518 0.0518 0.0548 0.0132 0.0002 0.8698
VaR0.05 13.0510 0.0573 0.0726 0.2069 0.0135 0.0002 0.8984
ES0.05 16.7847 0.0598 0.0804 0.2600 0.0135 0.0002 1.3299

Case 2: k = 1%
Mean of tran. costs 0.0733 1.0322 1.0324 1.0357 0.1572 0.1669 0.1117

Mean 0.2078 1.0046 1.0046 1.0127 0.1501 0.1683 0.1109

MAD 2.4134 0.6422 0.6445 0.6611 0.0725 0.0970 0.1827
E(HE 2 ) 35.8495 3.1795 3.1946 3.2696 0.0463 0.0657 0.2364
E(HE + ) 1.4327 1.0534 1.0534 1.0618 0.1523 0.1683 0.1903
VaR0.05 13.1242 4.6753 4.6940 4.6273 0.5232 0.5837 1.1738
ES0.05 16.8580 5.3307 5.3410 5.2645 0.5505 0.6091 1.6009

Case 3: k = 2%
Mean of tran. costs 0.1466 2.0644 2.0648 2.0714 0.3144 0.3338 0.2234

Mean 0.2811 2.0368 2.0370 2.0484 0.3074 0.3352 0.2226

MAD 2.4134 1.2622 1.2645 1.2830 0.1588 0.1927 0.1894
E(HE 2 ) 35.8853 12.2066 12.2378 12.4254 0.2000 0.2597 0.2908
E(HE + ) 1.5060 2.0749 2.0751 2.0866 0.3074 0.3352 0.2816
VaR0.05 13.1975 9.2897 9.3090 9.2589 1.0465 1.1564 1.4492
ES0.05 16.9313 10.6043 10.6155 10.5603 1.0943 1.2037 1.8719

Case 4: k = 3%
Mean of tran. costs 0.2199 3.0944 3.0972 3.1071 0.4716 0.5007 0.3351

Mean 0.3544 3.0690 3.0693 3.0841 0.4646 0.5020 0.3343

MAD 2.4134 1.8905 1.8929 1.9132 0.2484 0.2884 0.2022
E(HE 2 ) 35.9318 27.3279 27.3767 27.7176 0.4667 0.5820 0.3862
E(HE + ) 1.5791 3.1030 3.1033 3.1182 0.4646 0.5020 0.3751
VaR0.05 13.2708 13.9040 13.9239 13.8905 1.5757 1.7293 1.7246
ES0.05 17.0045 15.8779 15.8899 15.8561 1.6422 1.7984 2.1430

in transaction costs than the others, especially the MSS method. This phenomenon is due
to the number of components used to construct the hedging portfolio. For example, since
the MDEK100 method uses 201 securities to construct the portfolio for hedging the UOC
option, it replicates the UOC option better than does the MDEK6 method if k = 0. However,
from the mean transaction costs of each hedging method shown in Tables 1 and 2, we see
that the MDEK100 method suffers from transaction costs more than the MDEK6 method
and its adjusted hedging loss grows faster than that of the MDEK6 method as k increases.
Similar phenomena can be found for the NP method.

© 2015 Australian Statistical Publishing Association Inc.

Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

1467842x, 2015, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/anzs.12120 by University Rostock Universitätsbibliothek, Wiley Online Library on [26/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
S. F. HUANG AND C. Y. TSAI 313

Table 2
The hedging performances of the CC, NP0, NP4, NP9, MDEK6, MDEK100 and MSS methods
for UOC options with S0 = 100, r = 0.05, q = 0,  = 0.15, K = 100, T = 365 (days), B = 120
and m = 104 random paths under the BS model.

CC NP0 NP4 NP9 MDEK6 MDEK100 MSS

Case 1 : k = 0% UOC0 (B, K) = 2.1208

Mean 0.3871 −0.0289 −0.0287 −0.0257 −0.0093 −0.0000 −0.0115

MAD 4.6591 0.1062 0.1095 0.1292 0.0319 0.0004 0.4177
E(HE 2 ) 66.8904 0.3276 0.3286 0.3341 0.0062 0.0000 0.5714
E(HE + ) 9.6430 0.0743 0.0748 0.0826 0.0211 0.0003 1.0609
VaR0.05 16.1579 0.0841 0.1040 0.2950 0.0227 0.0004 1.5981
ES0.05 18.0371 0.0848 0.1095 0.3083 0.0227 0.0004 1.9443

Case 2 : k = 1%
Mean of tran. costs 0.1709 2.8043 2.8048 2.8168 0.3440 0.3703 0.2466

Mean 0.5580 2.7754 2.7761 2.7912 0.3389 0.3749 0.2351

MAD 4.6591 1.6433 1.6483 1.6692 0.1657 0.2097 0.4090
E(HE 2 ) 67.0520 12.8595 12.8880 12.9973 0.1586 0.2029 0.6187
E(HE + ) 3.0790 2.8450 2.8460 2.8617 0.3411 0.3749 0.4514
VaR0.05 16.3288 6.9167 6.9145 6.9199 0.6698 0.7287 1.9555
ES0.05 18.2080 7.2491 7.2436 7.2835 0.6845 0.7420 2.2983

Case 3 : k = 2%
Mean of tran. costs 0.3418 5.6086 5.6096 5.6336 0.6880 0.7406 0.4932

Mean 0.7289 5.5797 5.5809 5.6080 0.6829 0.7452 0.4818

MAD 4.6591 3.2559 3.2611 3.2886 0.3523 0.4152 0.4126
E(HE 2 ) 67.2719 50.5094 50.5747 51.0082 0.6492 0.7993 0.8137
E(HE + ) 3.2485 5.6303 5.6315 5.6588 0.6829 0.7452 0.6355
VaR0.05 16.4997 13.7600 13.7594 13.7991 1.3135 1.4378 2.3129
ES0.05 18.3789 14.4264 14.4226 14.4995 1.3364 1.4621 2.6522

Case 4 : k = 3%
Mean of tran. costs 0.5127 8.4129 8.4144 8.4504 1.0320 1.1109 0.7398

Mean 0.8999 8.3840 8.3857 8.4248 1.0269 1.1156 0.7284

MAD 4.6591 4.8805 4.8860 4.9200 0.5423 0.6207 0.4300
E(HE 2 ) 67.5503 113.2772 113.3887 114.3668 1.4780 1.7892 1.1564
E(HE + ) 3.4186 8.4294 8.4312 8.4705 1.0269 1.1156 0.8220
VaR0.05 16.6707 20.6034 20.6043 20.6783 1.9568 2.1469 2.6703
ES0.05 18.5498 21.6037 21.6015 21.7156 1.9901 2.1823 3.0061

We also conducted many other simulation scenarios by changing the values of the
parameters q, , K and B, with results similar to those presented in Tables 1 and 2. We do
not present these other results here to save space. In particular we found that if  increases
then the values of the components in each hedging portfolio increase, which leads to an
increase in the effect of the transaction costs on the performance of the hedging strategies.
In this situation, the superiority of the MSS method over the other strategies becomes
more pronounced. Consequently we recommend applying the MSS method for hedging
barrier options under the BS model if the volatility is large or if high transaction costs are
charged.

© 2015 Australian Statistical Publishing Association Inc.

Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

1467842x, 2015, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/anzs.12120 by University Rostock Universitätsbibliothek, Wiley Online Library on [26/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
314 HEDGING BARRIER OPTIONS IN GARCH MODELS

Table 3
The hedging performances of the NP0, NP4, NP9, MDEK6, MDEK100 and MSS methods
for UOC options with S0 = 100, r = 0.05, K = 110, T = 365 (days), B = 130
and m = 104 random paths under the GARCH model.

NP0 NP4 NP9 MDEK6 MDEK100 MSS

Case 1 : k = 0% UOC0 (B, K) = 1.2504

Mean −0.1054 −0.1091 −0.1001 −0.0910 −0.0913 0.1418

MAD 0.0500 0.1627 0.1192 0.2044 0.0309 0.2889
E(HE 2 ) 0.0946 0.2545 0.1560 0.3354 0.0206 0.4091
E(HE + ) 0.0000 0.3746 0.2083 0.5988 0.0623 1.1262
VaR0.05 −0.0568 0.4079 0.2492 0.6091 −0.0429 1.6758
ES0.05 −0.0423 0.6476 0.4475 1.1062 0.0139 2.1508

Case 2 : k = 1%
Mean of tran. costs 2.5089 2.6187 2.6602 0.3063 0.4860 0.1785

Mean 2.4035 2.5096 2.5601 0.2117 0.3946 0.3205

MAD 1.5591 1.6182 1.6793 0.2777 0.2859 0.3048
E(HE 2 ) 12.1990 13.5144 14.0763 0.4558 0.3840 0.5338
E(HE + ) 2.4228 2.5244 2.5823 0.3752 0.3972 0.3866
VaR0.05 7.6970 8.1746 8.1733 1.3251 1.4533 1.9881
ES0.05 8.3012 8.8052 9.2495 1.8341 1.6382 2.4575

Case 3 : k = 2%
Mean of tran. costs 5.0178 5.2374 5.3204 0.6126 0.9720 0.3570

Mean 4.9124 5.1282 5.2203 0.5143 0.8806 0.5014

MAD 3.1277 3.2630 3.3450 0.4129 0.5874 0.3391
E(HE 2 ) 49.6617 54.6234 56.8634 0.9214 1.7069 0.7551
E(HE + ) 4.9260 5.1391 5.2351 0.5539 0.8812 0.5309
VaR0.05 15.4926 16.3753 16.2950 2.0732 2.9993 2.3157
ES0.05 16.6834 17.6902 18.3661 2.5785 3.3777 2.7725

Case 4 : k = 3%
Mean of tran. costs 7.5267 7.8561 7.9806 0.9189 1.4580 0.5355

Mean 7.4212 7.7469 7.8805 0.8169 1.3665 0.6799

MAD 4.7024 4.9099 5.0179 0.5723 0.8909 0.3905
E(HE 2 ) 112.4830 123.5815 128.5171 1.7324 3.9894 1.0622
E(HE + ) 7.4291 7.7547 7.8886 0.8411 1.3669 0.6843
VaR0.05 23.2929 24.5879 24.5054 2.8446 4.5571 2.6284
ES0.05 25.0655 26.5858 27.5073 3.3379 5.1249 3.0791

4.2. Hedging performance under the GARCH model

In the following, we compare the hedging performance of the NP4, NP9, MDEK6,
MDEK100 and MSS methods under the GARCH model defined by (22), where S0 = 100,
r = 0.1 (annual), T = 365 (days), 0 = 0.00001, 1 = 0.2, 2 = 0.7, = 0.01, and 12 =
−1
0 (1 − 1 − 2 ) , these being the paramater values that were used in Duan & Simonato
(2001). Tables 3 and 4 present the simulation results with m = 104 random paths, k = 0%,
1%, 2% and 3%, and (K, B) = (110, 130) and (100, 120), respectively. From Tables 3 and
4 we see that the MDEK100 and NP0 methods have better hedging performance than the
others if the transaction cost is neglected. Conversely if transaction costs are present, the

© 2015 Australian Statistical Publishing Association Inc.

Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

1467842x, 2015, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/anzs.12120 by University Rostock Universitätsbibliothek, Wiley Online Library on [26/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
S. F. HUANG AND C. Y. TSAI 315

Table 4
The hedging performances of the NP0, NP4, NP9, MDEK6, MDEK100 and MSS methods
for UOC options with S0 = 100, r = 0.05, K = 100, T = 365 (days), B = 120
and m = 104 random paths under the GARCH model.

NP0 NP4 NP9 MDEK6 MDEK100 MSS

Case 1 : k = 0% UOC0 (B, K) = 1.5237

Mean −0.1588 −0.1932 −0.2550 −0.1411 −0.1505 0.2179

MAD 0.0863 0.4039 0.3336 0.2904 0.0323 0.4432
E(HE 2 ) 0.2036 0.9834 0.8869 0.4678 0.0302 0.6572
E(HE + ) 0.0000 0.7108 0.9792 0.5135 0.0647 1.1557
VaR0.05 −0.0446 0.4439 0.3207 0.7077 −0.0793 2.0776
ES0.05 −0.0344 2.2990 1.9273 1.1575 −0.0178 2.5131

Case 2 : k = 1%
Mean of tran. costs 4.7270 5.5023 6.4794 0.4829 0.8521 0.3127

Mean 4.5682 5.3091 6.2244 0.4443 0.7016 0.5306

MAD 2.7281 3.1331 3.7216 0.4314 0.4817 0.4618
E(HE 2 ) 30.1763 40.6779 56.1238 0.7528 0.7787 0.9325
E(HE + ) 4.6014 5.3248 6.2383 0.5540 0.7037 0.6412
VaR0.05 9.3248 10.8414 12.9219 1.6643 1.5956 2.4740
ES0.05 9.6233 11.5216 13.9723 2.0675 1.7303 2.9046

Case 3 : k = 2%
Mean of tran. costs 9.4540 11.0046 12.9588 0.9658 1.7042 0.6254

Mean 9.2952 10.8115 12.7037 1.0296 1.5537 0.8434

MAD 5.4519 6.3112 7.4690 0.6886 0.9706 0.5128
E(HE 2 ) 122.6828 165.7755 230.0631 1.9898 3.5643 1.4353
E(HE + ) 9.3162 10.8179 12.7078 1.0884 1.5543 0.8573
VaR0.05 18.7092 21.8481 26.1008 2.6513 3.3588 2.8704
ES0.05 19.3387 22.9764 27.6592 3.0176 3.5953 3.2961

Case 4 : k = 3%
Mean of tran. costs 14.1810 16.5069 19.4382 1.4487 2.5563 0.9381

Mean 14.0222 16.3138 19.1831 1.6149 2.4057 1.1561

MAD 8.1845 9.4975 11.2181 0.9855 1.4597 0.6035
E(HE 2 ) 277.7233 376.2762 522.7050 4.1787 8.3867 2.1658
E(HE + ) 14.0300 16.3171 19.1831 1.6461 2.4060 1.1561
VaR0.05 28.0917 32.9264 39.2454 3.6312 5.1262 3.2668
ES0.05 29.0544 34.5165 41.3994 4.0064 5.4620 3.6877

results indicate that the MSS method has better hedging performance than the other methods,
especially when k = 3%. This phenomenon is consistent with what we have found in the
case of the BS model. In addition note that the annualized stationary volatility in the context
of this simulation case is around 0.60, which is greater than the corresponding  = 0.15 in
Tables 1 and 2. As mentioned at the end of last section, comparing results in Tables 1–4
indicates that the superiority of the MSS method over the other strategies is more substantial
when volatility is large.
To provide more insight into the hedging performance of the NP, MDEK and MSS
methods under the GARCH framework, we have plotted the hedging losses of each method

© 2015 Australian Statistical Publishing Association Inc.

Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

1467842x, 2015, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/anzs.12120 by University Rostock Universitätsbibliothek, Wiley Online Library on [26/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
316 HEDGING BARRIER OPTIONS IN GARCH MODELS

NP0 method
6

4
hedging loss

0
B = 120
−2
B = 120 with tran. cost
−4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
τ

10

5
hedging loss

−5 B = 130
B = 130 with tran. cost
−10
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
τ

20

10
hedging loss

−10 B = 140
B = 140 with tran. cost
−20
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
τ

Figure 1. Plots of the hedging losses of the NP0 method versus  with zero (solid lines) and 1%
(dashed lines) transaction costs, respectively, under the GARCH model. The upper, middle, and lower
panels correspond to barrier prices of 120, 130 and 140 respectively.

against the first passage time . Figures 1–4 display the hedging loss of the NP0, NP4,
MDEK6 and MSS methods, respectively, against different values of  with B = 120, 130 and
140. In these figures the solid lines correspond to the hedging losses under zero transaction
costs and the dashed lines correspond to the case of k = 1%. We see from these figures that
the transaction costs did indeed enlarge the hedging losses of each method for each  since
the dashed lines are above the corresponding solid lines. In particular, Figure 1 indicates that
the pure strike-spread NP0 method has a very stable replicable performance under the UOC
option if k = 0 because most of the hedging losses of the solid lines are zero. It also provides
the trader with positive gains when  approaches the maturity date since the hedging losses
are negative when  is large. However the effect of transaction costs on the NP0 method
is greater than on the MDEK and MSS methods since the increment between the dashed
lines and solid lines of the NP0 is dramatically larger than the other two methods. When a
trader closes out her hedging portfolio at time , she has to pay transaction fees for both
long and short positions in the portfolio. A large number of long and short positions leads
to an increase in the overall transaction costs of the hedging procedure. Consequently the
severe impact of the transaction costs on the NP0 indicates that the NP0 method entails

© 2015 Australian Statistical Publishing Association Inc.

Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

1467842x, 2015, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/anzs.12120 by University Rostock Universitätsbibliothek, Wiley Online Library on [26/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
S. F. HUANG AND C. Y. TSAI 317

NP4 method
10
hedging loss

0
B = 120
B = 120 with tran. cost
−5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
τ

15

10
hedging loss

−5 B = 130
B = 130 with tran. cost
−10
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
τ

15

10
hedging loss

−5 B = 140
B = 140 with tran. cost
−10
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
τ

Figure 2. Plots of the hedging losses of the NP4 method versus  with zero (solid lines) and 1%
(dashed lines) transaction costs, respectively, under the GARCH model. The upper, middle, and lower
panels correspond to barrier prices of 120, 130 and 140 respectively.

having larger numbers of long and short positions in its hedging portfolio than have the
MDEK and MSS methods. Similar phenomena can be discerned from Figure 2 for the
NP4 method except that the NP4 method fluctuates more than the NP0 method since it
involves a calendar-spread component in the hedging portfolio. Likewise Figure 3 displays
the fluctuation of the hedging loss of the modified calendar-spread MDEK6 method. Figure
4 indicates that the MSS method uniformly suffers less from transaction costs than the other
methods. Consequently the MSS method is recommended to traders when high transaction
costs are charged.

5. Empirical study
In this section we investigate the hedging performance of the proposed MSS method
and the MDEK method with n = 6 applied to a real data set. The spot prices of the daily
USD/EUR exchange rate from 1 January 2004 to 30 September 2006 (1004 observations)
were downloaded from http://www.oanda.com/lang/cnt/currency/historical-rates/ and were
fitted to the following GARCH model:

© 2015 Australian Statistical Publishing Association Inc.

Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

1467842x, 2015, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/anzs.12120 by University Rostock Universitätsbibliothek, Wiley Online Library on [26/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
318 HEDGING BARRIER OPTIONS IN GARCH MODELS

MDEK method
1

0.5
hedging loss

−0.5

−1

−1.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
τ

1
hedging loss

−1

−2

−3
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
τ

2
hedging loss

−2

−4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
τ

Figure 3. Plots of the hedging losses of the MDEK6 method versus  with zero (solid lines) and
1% (dashed lines) transaction costs, respectively, under the GARCH model. The upper, middle, and
lower panels correspond to barrier prices of 120, 130 and 140 respectively.

Rt = rUSD − rEUR + t − 0.5t2 + t "t ,

(27)
t2 = 0 + 1 t−1 "t−1 + 2 t−1 .
2 2 2

In (27) Rt = log(St =St−1 ) is the log return of the USD/EUR exchange rate, St , at time t and "t
are i.i.d. N (0, 1) random variables. The fitted values of the parameters are ˆ = 0.0081,
ˆ 0 = 2 × 10−7 , ˆ 1 = 0.0196 and ˆ 2 = 0.9705, with T = 365 (days), rUSD = 6.38 × 10−5
(daily) and rEUR = 5.69 × 10−5 (daily). The parameters rUSD and rEUR are the averages
of the yields in USD and EUR, respectively, obtained from the data set introduced be-
low. Figure 5 presents a plot of market exchange rates (dots) and GARCH fitted values
(line) against time and a scatter plot of the market exchange rates versus the GARCH
fitted values. From Figure 5 we see that the GARCH fitted values obtained from model
(27) provide accurate estimates of the market values of the daily USD/EUR exchange
rate.
The data on barrier option prices covering the period 2 January 2004 to 27 September
2005 come from the risk-management department of Danske Bank and a detailed description
of the data set can be found in Jessen & Poulsen (2013). We considered only UOC barrier
options and excluded options with values lower than 10−5 in our evaluation of the hedging

© 2015 Australian Statistical Publishing Association Inc.

Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

1467842x, 2015, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/anzs.12120 by University Rostock Universitätsbibliothek, Wiley Online Library on [26/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
S. F. HUANG AND C. Y. TSAI 319

MSS method
3
B = 120
2 B = 120 with tran. cost
hedging loss

−1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
τ

3
B = 130
2
B = 130 with tran. cost
hedging loss

−1

−2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
τ

4
B = 140
2 B = 140 with tran. cost
hedging loss

−2

−4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
τ

Figure 4. Plots of the hedging losses of the MSS method versus  with zero (solid lines) and 1%
(dashed lines) transaction costs, respectively, under the GARCH model. The upper, middle, and lower
panels correspond to barrier prices of 120, 130 and 140 respectively.

performance. In order to be able to separate the lifetime of the UOC option into six time
intervals for the MDEK6 method we included only those UOC options having at least
30 days to expiry in our investigation. This left us with a total of 292 observations. We
used the yields in USD and EUR of these 292 observations to compute the values of the
parameters rUSD and rEUR previously mentioned. Since we did not have intraday observations
of the USD/EUR exchange rate, we made use of the daily spot prices of the USD/EUR
exchange rate to determine when the first passage time occurs. We found that 46 of the
292 observations passed their corresponding barrier prices before expiry. In the process of
constructing the MSS and MDEK6 hedging strategies at the initial time and to compute
the hedging loss at the first passage time, the values of standard call, binary, and UID
options at the initial or the first passage times in (12), (15), (17) and (21) were computed
by the EPMS method of Huang (2014) and Huang & Tu (2014) under the GARCH model
(27).
We used the same six measures, that were used to assess hedging performance in
Section 1, in the current setting. Note that here HEj denotes the adjusted hedging loss with
transaction costs for the jth observation, j = 1,…, 292. Table 5 presents the values of the six

© 2015 Australian Statistical Publishing Association Inc.

Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

1467842x, 2015, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/anzs.12120 by University Rostock Universitätsbibliothek, Wiley Online Library on [26/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
320 HEDGING BARRIER OPTIONS IN GARCH MODELS

(a) Time plot of market values vs. GARCH estimates

1.5

1.4

1.3

1.2

1.1
50 100 150 200 250 300 350 400 450 500

1.5

1.4

1.3

1.2

1.1
550 600 650 700 750 800 850 900 950 1000

(b) Scatter plot of market values vs. GARCH estimates

1.35
Market values

1.3
1.25
1.2
1.15
1.15 1.2 1.25 1.3 1.35 1.4
GARCH estimates

Figure 5. (a) Plots of the market exchange rates (dots) and the GARCH fitted values (solid lines)
from 1 January 2004 to 30 September 2006. (b) Plot of the market exchange rates versus the GARCH
fitted values.

measures for the MSS and MDEK6 methods with k = 0, 1% and 3%. In Table 5 most of the
values of the six measures of the MSS method are smaller than those of the MDEK6 method,
especially when transaction costs are larger. This phenomenon is consistent with what we
showed in the simulation study and provides yet more justification for recommending the
proposed MSS method when high transaction costs are charged.

6. Conclusion
This study proposes a MSS method for hedging UOC options under the BS and GARCH
models and compares its hedging performance with the pure strike-spread method of Carr &
Chou (1997), the unified method of Nalholm & Poulsen (2006) and the modified calendar-
spread method of Chung et al. (2010). An accurate, easy-to-implement and fast procedure for
generating the first passage time for GARCH dynamics is established, which facilitates a fair
comparison among the various hedging methods. Simulation results indicate that the MSS and
MDEK6 methods suffer less from large transaction costs than the other hedging strategies.
Numerical results also indicate that the superiority of the MSS method over the other strategies
is more substantial when volatility is large. Furthermore we present an empirical study that
compares the hedging performance of the MSS and MDEK6 methods on the basis of a
real data set. The results of this comparison indicate that the MSS method performs better
than does the MDEK6, especially when transaction costs are larger. Consequently the MSS

© 2015 Australian Statistical Publishing Association Inc.

Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

1467842x, 2015, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/anzs.12120 by University Rostock Universitätsbibliothek, Wiley Online Library on [26/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
S. F. HUANG AND C. Y. TSAI 321

Table 5
The hedging performances of the MDEK6 and MSS methods for
UOC options during 2 January 2004 to 27 September 2005 under
the GARCH model.

k =0 k = 1% k = 3%

MDEK6 MSS MDEK6 MSS MDEK6 MSS

Mean 0.0006 −0.0007 0.0008 −0.0003 0.0015 0.0004
MAD 0.0006 0.0007 0.0008 0.0006 0.0010 0.0005
E(HE 2 ) 1×10−6 4×10−6 2×10−6 2×10−6 4×10−6 9×10−7
E(HE + ) 0.0009 0 0.0011 0.0002 0.0015 0.0007
VaR0.05 0.0025 0 0.0037 0.0004 0.0046 0.0013
ES0.05 0.0031 0 0.0043 0.0005 0.0053 0.0014

approach is recommended for hedging barrier options when the volatility is large or high
transaction costs are charged.

Appendix A: The derivation of the CC hedge for the UOC option

By put-call parity we have UOC0 (B, K) = C0 (K, T ) − UIC0 (B, K), where UIC0 (B, K) =
(ST − K)+ I(max0tT St B) is an up-and-in call option (UIC). Thus a UOC can be hedged by
holding a long position of a European call and a short position of a UIC. According to results
due to Reiner & Rubinstein (1991), the valuation formula of a UIC option with strike price
K is
B 2(r−q)
UICT (B, K) = Be−qT ( ) 2 ((d5 ) − (d1 )) − Ke−rT (d4 )
S
B 2(r−q)
− e K( ) 2 −1 ((d6 ) − (d2 )) + e−qT S(d3 ),
−rT
S
2
log(Gi )+(r−q+ 2 )T √ B2
where di = √
 T
, di+1 = di −  T , i = 1, 3, 5, G1 = BS , G3 = BS , G5 = KS , and
(·) is the cumulative distribution function of a standard normal distribution. Removing
the requirement that B > S and letting T → 0 yield the following adjusted payoff of a UIC
option:


f (S) = lim UIC(B, K)
T →0
2 (28)
B S
=( − K)( )p [I(S< B2 ) − I(S<B) ] + (S − K)I(S>B) .
S B K

In (28) p = 1 − 2(r−q)
2
. By (1) with  = B− and (28), it suffices to compute 
f (B− ), 
f  (B− ) and

f  (h) to form the CC hedge for a UIC option. Let (·) be the Fermi-Dirac delta function,
which is a generalized function or distribution on the real line (Dirac
 ∞1958) and is defined by
the following properties: (x) = ∞ if x = 0, (x) = 0 if x = 0, and −∞ (x)dx = 1. The delta
function is sometimes called the impulse symbol (Bracewell 2000, pp. 74–104) and can be
x2
viewed as a limit of a Gaussian (x) = lim→0 √2
1
e− 22 or a Lorentzian (x) = lim→0 1 
 x2 +2
.
By straightforward computation and using the fact that x (x) = 0, we have

© 2015 Australian Statistical Publishing Association Inc.

Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

1467842x, 2015, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/anzs.12120 by University Rostock Universitätsbibliothek, Wiley Online Library on [26/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
322 HEDGING BARRIER OPTIONS IN GARCH MODELS

 p S B2 S
f  (S) = [ ( )p−1 ( − K) − ( )p−2 ][I(S< B2 ) − I(S<B) ]
B B S B K
(29)
S p B2
+ [( ) ( − K) + S − K] (S − B) + I(S>B)
B S
and

 S p − 2 pK S B2
f  (S) = ( )p−2 (p − 1)[ − 2 ]I(BS< B2 ) + ( )p−2 (S − )
B S B K B K
S p−2 Sp B2 B p−2
+ 2( ) [ 2 ( − K) − 1 + ( ) ] (S − B) (30)
B B S S
S p B2
+ [( ) ( − K) + S − K]  (S − B).
B S

In (30)  (·) denotes the first order derivative of (·). By (29), (30) and the assumption that
B > K, we have  f (B− ) = 0, 
f  (B− ) = 0 and
⎧  (h)I 2 + 2 (h) (h −
B2
)
⎪
⎨
1 (BS< BK ) K

f (h) = +3 (h) (h − B) + 4 (h)  (h − B), for h   = B− , (31)
⎪
⎩
4 (h)  (h − B), for h <  = B− .
2
In (31) 1 (h) = ( Bh )p−1 (p − 1)[ p−2 h
− pK
B2
], 2 (h) = ( Bh )p−2 , 3 (h) = 2( Bh )p−2 [ Bhp2 ( Bh − K) − 1 +
2
( Bh )p−2 ] and 4 (h) = [( Bh )p ( Bh − K) + h − K].
By using UOC0 (B, K) = C0 (K, T ) − UIC0 (B, K) and substituting (31) into (1), we obtain
the following CC hedge for the UOC option:
 B−
V0 = C0 (K, T ) − 4 (h)  (h − B)P0 (h, T )dh
0
 B2
K
− 1 (h)C0 (h, T )dh
B
 (32)
∞
B2
− {2 (h) (h − ) + 3 (h) (h − B)}C0 (h, T )dh
− K
B∞
− 4 (h)  (h − B)C0 (h, T )dh.
B−
∞ ∞ 
From the identities
∞  −∞ f (s) (s − x)ds = f (s), x (x) = 0 and −∞ (x) (x)dx
= − −∞  (x) (x)dx we find that
 B−  B−

4 (h) (h − B)P0 (h, T )dh = − [4 (h)P0 (h, T ) + 4 (h)P0 (h, T )] (h − B)dh = 0,
0 0
(33)
 ∞
B2 B B2 p
{2 (h) (h − ) + 3 (h) (h − B)}C0 (h, T )dh = ( )p−2 C0 ( , T ) + 2 (B − K)C0 (B, T )
B− K K K B
(34)

© 2015 Australian Statistical Publishing Association Inc.

Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

1467842x, 2015, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/anzs.12120 by University Rostock Universitätsbibliothek, Wiley Online Library on [26/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
S. F. HUANG AND C. Y. TSAI 323

and
 ∞
4 (h)  (h − B)C0 (h, T )dh
B−
 ∞
=− [4 (h)C0 (h, T ) + 4 (h)C0 (h, T )] (h − B)dh (35)
B−
p
= 2(B − K)Bin0 (B, T ) − (B − K)C0 (B, T ).
B

In (34) and (35) C0 (h, T ) = S0 e−qT (1 ) − he−rT (2 ) and C0 (h, T ) = @C0@h (h, T )
. In turn this lat-
−rT −rT
ter partial derivative is equal to −e (2 ) = −Bin0 (h, T ) since he (2 ) = S0 e−qT (1 ).
√
In these last expressions 1 = log(S0 =h)+(r−q+0.5
2
√
 T
)T
, 2 = 1 −  T and Bin0 (h, T ) = e−rT (2 )
is the initial price of a binary call option having payoff I(ST >h) at the expiration date T .
By (32)–(35) 2the initial hedging portfolio for the CC method of the UOC option is V0 =
B 2
C0 (K, T ) − BK 1 (h)C0 (h, T )dh − ( KB )p−2 C0 ( BK , T ) − Bp (B − K)C0 (B, T ) − 2(B − K)Bin0 (B).
Hence (2) follows.

References
Atiya, A.F. & Metwally, S.A. (2005). Efficient estimation of first passage time density function for
jump-diffusion processes. SIAM J. Sci. Comput. 26, 1760–1775.
Bollerslev, T. (1986). Generalized autoregressive conditional heteroskedasticity. J. Econ. 31, 307–327.
Bowie, J. & Carr, P. (1994). Static simplicity. Risk 7, 45–49.
Bracewell, R. (2000). The Fourier Transform and Its Applications, 3rd edn. New York: McGraw-Hill.
Carr, P. & Chou, A. (1997). Breaking barriers: Static hedging of barrier securities. Risk 10, 139–144.
Carr, P. & Lee, R. (2009). Put-call symmetry: Extensions and applications. Math. Finance 19, 523–560.
Carr, P., Ellis, K. & Gupta, V. (1998). Static hedging of exotic option. J. Finance 53, 1165–1190.
Chung, S.L., Shih, P.T. & Tsai, W.C. (2010). A modified static hedging method for continuous barrier
options. J. Futures Mark. 30, 1150–1166.
Derman, E., Ergener, D. & Kani, I. (1995). Static options replication. J. Deriv. 2, 78–95.
Dirac, P. (1958). The Principles of Quantum Mechanics, 4th edn. Oxford: Clarendon Press.
Duan, J.C. (1995). The GARCH option pricing model. Math. Finance 5, 13–32.
Duan, J.C. & Simonato, J.G. (1998). Empirical martingale simulation for asset prices. Manag. Sci. 44,
1218–1233.
Duan, J.C. & Simonato, J.G. (2001). American option pricing under GARCH by a Markov chain approx-
imation. J. Econom. Dynam. Control 25, 1689–1718.
Dupont, D.Y. (2001). Hedging barrier options: Current methods and alternatives. Technical Report, The
Netherlands: EURANDOM-TUE.
Engelmann, B., Fengler, M.R., Nalholm, M. & Schwendner, P. (2006). Static versus dynamic
hedges: An empirical comparison for barrier options. Rev. Deriv. Res. 9, 239–264.
Engle, R. (1982). Autoregressive conditional heteroskedasticity with estimates of the variance of United
Kingdom inflation. Econometrica 50, 987–1008.
Engle, R. & Rosenberg, J. (1994). Hedging options in a GARCH environment: Testing the term structure
of stochastic volatility models. NBER Working Paper Series no. 4958.
Engle, R. & Rosenberg, J. (1995). GARCH gamma. J. Deriv. 2, 47–59.
Heston, S.L. & Nandi, S. (2000). A closed-form GARCH option valuation model. Rev. Financ. Stud. 13,
585–625.
Huang, S.F. (2014). A modified empirical martingale simulation for financial derivative pricing. Comm.
Statist. Theory Methods 43, 328–342.
Huang, S.F. & Guo, M.H. (2009). Financial derivative valuation – A dynamic semiparametric approach.
Statist. Sinica 19, 1037–1054.
Huang, S.F. & Tu, Y.T. (2014). Asymptotic distribution of the EPMS estimator for financial derivatives
pricing. Comput. Statist. Data Anal. 73, 129–145.

© 2015 Australian Statistical Publishing Association Inc.

Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

1467842x, 2015, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/anzs.12120 by University Rostock Universitätsbibliothek, Wiley Online Library on [26/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
324 HEDGING BARRIER OPTIONS IN GARCH MODELS

Hull, J. (2012). Options, Futures, and Other Derivatives, 8th edn. Boston: Prentice Hall.
Jessen, C. & Poulsen, R. (2013). Empirical performance of modles for barrier option valuation. Quant.
Finance 13, 1–13.
Kallsen, J. & Taqqu, M. (1998). Option pricing in ARCH-type models. Math. Finance 8, 13–26.
Nalholm, M. & Poulsen, R. (2006). Static hedging of barrier options under general asset dynamics:
unification and application. J. Deriv. 13, 46–60.
Reiner, E. & Rubinstein, M. (1991). Breaking down the barriers. Risk 4, 28–35.
Tsay, R.S. (2010). Analysis of Financial Time Series, 3rd edn. New Jersey: Wiley.

© 2015 Australian Statistical Publishing Association Inc.