Vanna Volga FX
Vanna Volga FX
call
= DF
f
(t, T) N(d
1
)
put
= DF
f
(t, T) N(d
1
) (2)
Setting up the corresponding Delta hedge will make ones position insensitive to small FX spot move-
ments if one is measuring risks in a USD (domestic) risk-neutral world. Other currency pairs (e.g.
USDJPY) use the premium included Delta convention:
call
=
K
S
DF
d
(t, T) N(d
2
)
put
=
K
S
DF
d
(t, T) N(d
2
) (3)
The quantities (2) and (3) are expressed in Ccy1, which is by convention the unit of the quoted Delta.
Taking the example of USDJPY, setting up the corresponding Delta hedge (3) will make ones position
insensitive to small FX spot movements if one is measuring risks in a USD (foreign) risk-neutral world.
1
for a description of our notation, see A.
4
Note that (2) and (3) are linked through the options premium (1), namely S(
call
call
) = Call
and similarly for the put (see B for more details).
With regards to the dependency on maturity, the so-called G11 currency pairs use a spot Delta
convention (2), (3) for short maturities (typically up to 1 year) while for longer maturities where the
interest rate risk becomes signicant, the forward Delta (or driftless Delta) is used, as the derivative
of the undiscounted premium with respect to forward:
F
call
= N(d
1
)
F
put
= N(d
1
)
F
call
=
K
S
DF
d
(t, T)
DF
f
(t, T)
N(d
2
)
F
put
=
K
S
DF
d
(t, T)
DF
f
(t, T)
N(d
2
)
(4)
where, as before, by tilde we denoted the premium-included convention. The Deltas in the rst row
represent the nominals of the forward contracts to be settled if one is to forward hedge the Delta risk
in a domestic currency while those of the second row consider a foreign risk neutral world. Other
currency pairs (typically those where interest-rate risks are substantial, even for short maturities) use
the forward Delta convention for all maturity pillars.
3.2 At-The-Money Conventions
As in the case of the Delta, the at-the-money (ATM) volatilities quoted by brokers can have various
interpretations depending on currency pairs. The ATM volatility is the value from the smile curve
where the strike is such that the Delta of the call equals, in absolute value, that of the put (this strike
is termed ATM straddle or ATM delta neutral ). Solving this equality yields two possible solutions,
depending on whether the currency pair uses the Black-Scholes Delta or the premium included Delta
convention. The 2 solutions respectively are:
K
ATM
= F exp
_
1
2
2
ATM
_
K
ATM
= F exp
_
1
2
2
ATM
_
(5)
Note that these expressions are valid for both spot and forward Delta conventions.
3.3 Smile-related quotes and the brokers Strangle
Let us assume that a smile surface is available as a function of the strike (K). In liquid FX markets
some of the most traded strategies include
Strangle(K
c
, K
p
) = Call(K
c
, (K
c
)) + Put(K
p
, (K
p
)) (6)
Straddle(K) = Call(K,
ATM
) + Put(K,
ATM
) (7)
Buttery(K
p
, K, K
c
) =
1
2
_
Strangle(K
c
, K
p
) Straddle(K)
(8)
Brokers normally quote volatilities instead of the direct prices of these instruments. These are
expressed as functions of , for instance a volatility at 25-call or put refers to the volatility at the
strikes K
c
, K
p
that satisfy
call
(K
c
, (K
c
)) = 0.25 and
put
(K
p
, (K
p
)) = 0.25 respectively (with
the appropriate Delta conventions, see section 3.1). Typical quotes for the vols are
at-the-money (ATM) volatility:
ATM
25-Risk Reversal (RR) volatility:
RR25
5
1-vol-25-Buttery (BF) volatility:
BF25(1vol)
2-vol-25-Buttery (BF) volatility:
BF25(2vol)
By market convention, the RR vol is interpreted as the dierence between the call and put implied
volatilities respectively:
RR25
=
25C
25P
(9)
where
25C
= (K
c
) and
25P
= (K
p
).
The 2-vol-25-Buttery can be interpreted through
BF25(2vol)
=
25C
+
25P
2
ATM
(10)
Associated to the
BF25(2vol)
is the 2-vol-25-strangle vol dened through
STG25(2vol)
=
BF25(2vol)
+
ATM
.
The 2-vol-25-Buttery value
BF25(2vol)
is in general not directly observable in FX markets.
Instead, brokers usually communicate the
BF25(1vol)
, using a brokers strangle or 1vol strangle con-
vention. The exact interpretation of
BF25(1vol)
can be explained in a few steps:
Dene
STG25(1vol)
=
ATM
+
BF25(1vol)
.
Solve equations (2),(3) to obtain K
25C
and K
25P
, the strikes where the Delta of a call is ex-
actly 0.25, and the Delta of a put is exactly -0.25 respectively, using the single volatility value
STG25(1vol)
.
Provided that the smile curve (K) is correctly calibrated to the market, then the quoted value
BF25(1vol)
is such that the following equality holds:
Call(K
25C
,
STG25(1vol)
) + Put(K
25P
,
STG25(1vol)
) = Call(K
25C
, (K
25C
)) + Put(K
25P
, (K
25P
))(11)
The dierence between
BF25(1vol)
and
BF25(2vol)
can be at times confusing. Often for convenience
one sets
BF25(2vol)
=
BF25(1vol)
as this greatly simplies the procedure to build up a smile curve.
However it leads to errors when applied to a steeply skewed market. Figure1 provides a graphical
interpretation of the quantities
STG25(1vol)
,
STG25(2vol)
,
BF25(1vol)
and
BF25(2vol)
in 2 very dierent
market conditions; the lower panel corresponds to the USDCHF-1Y smile, characterized by a relatively
mild skew, the upper panel corresponding to the extremely skewed smile of USDJPY-1Y. As a rule
of thumb one sets
BF25(2vol)
=
BF25(1vol)
when
RR25
is small in absolute value (typically < 1%).
When this empirical condition is not met,
BF25(1vol)
and
BF25(2vol)
represent actually two dierent
quantities, and substituting one for the other in the context of a smile construction algorithm would
yield substantial errors.
Table 1 gives more details about the numerical values used to produce the 2 smiles of Figure 1.
Various dierences are observed between the 2 smiles. In the USDCHF case, the values
BF25(2vol)
and
BF25(1vol)
are close to each other. Similarly, the strikes used in the 1vol-25 Strangle are rather
close to those attached to the 2vol-25 Strangle. On the contrary, in the USDJPY case, large dif-
ferences are observed between the parameters of the 1vol-25-Strangle and those of the 2vol-25-
Strangle.
Unfortunately, there is no direct mapping between
BF25(1vol)
and
BF25(2vol)
. This is mainly due
to the fact that these two instruments are attached to dierent points of the implied volatility curve.
The relationship between
BF25(2vol)
and
BF25(1vol)
implicitly depends on the entire smile curve.
6
Figure 1: Comparison between
STG25(2vol)
and
STG25(1vol)
, also called broker strangle in two dier-
ent smile conditions.
USDCHF USDJPY
date 8 Jan 09 28 Nov 08
FX spot rate 1.0902 95.47
maturity 1 year
r
d
1.3% 1.74%
r
f
2.03% 3.74%
ATM
16.85% 14.85%
RR25
-1.3% -9.4%
BF25(2vol)
1.1% 1.45%
BF25(1vol)
1.04% 0.2%
K
25P
/ K
25C
0.9586/1.2132 82.28/101.25
K
25P
/ K
25C
0.9630/1.2179 85.24/103.53
Table 1: Details of market quotes for the two smile curves of Figure 1.
In practice however, one may be interested in nding the value of
BF25(1vol)
from an existing smile
curve; this can be achieved using an iterative procedure:
pseudo-algorithm 1
1. Select an initial guess for
BF25(1vol)
2. compute the corresponding strikes K
25P
and K
25C
3. assess the validity of equality (11): compare the value of the Strangle (i) valued with a unique
vol
BF25(1vol)
(ii) valued with 2 implied vol corresponding to K
25P
, respectively K
25C
7
4. If the dierence between the two values exceeds some tolerance level, adapt the value
BF25(1vol)
and go back to 2.
In case one is given a value of
BF25(1vol)
from the market, and wants to use it to build an implied
smile curve, one may proceed the following way:
pseudo-algorithm 2
1. Select an initial guess for
BF25(2vol)
2. Construct an implied smile curve using
BF25(2vol)
and market value of
RR25
3. Compute the value of
BF25(1vol)
(for instance following guidelines of pseudo-algorithm 1)
4. Compare
BF25(1vol)
you obtained in 3 to the market-given one.
5. If the dierence between the two values exceeds some tolerance, adapt the value
BF25(2vol)
and
go back to 2.
To close this section on the brokers Strangle issue, let us clarify another enigmatic concept of FX
markets often used by practitioners, the so-called Vega-weighted Strangle quote. This is in fact an
approximation for the value of
STG25(1vol)
. To show this, we start from equality (11). First we assume
K
25P
= K
25P
and K
25C
= K
25C
. Next, we develop both sides in a rst order Taylor expansion in
around
ATM
. After canceling repeating terms on the left and right-hand side, we are left with:
(
STG25(1vol)
ATM
) (1(K
25P
,
ATM
) +1(K
25C
,
ATM
))
(
25P
ATM
) 1(K
25P
,
ATM
) + (
25C
ATM
) 1(K
25C
,
ATM
)
(12)
where 1(K, ) represents the Vega of the option, namely the sensitivity of the option price P with
respect to a change of the implied volatility: 1 =
P
STG25(1vol)
25P
1(K
25P
,
ATM
) +
25C
1(K
25C
,
ATM
)
1(K
25P
,
ATM
) +1(K
25C
,
ATM
)
(13)
which corresponds to the average (weighted by Vega) of the call and put implied volatilities.
Note that according to Castagna et al. [5] practitioners also use the term Vega-weighted buttery
for a structure where a strangle is bought and an amount of ATM straddle is sold such that the overall
vega of the structure is zero.
4 The Vanna-Volga Method
The Vanna-Volga method consists in adjusting the Black-Scholes TV by the cost of a portfolio which
hedges three main risks associated to the volatility of the option, the Vega, the Vanna and the Volga.
The Vanna is the sensitivity of the Vega with respect to a change in the spot FX rate: Vanna =
V
S
.
Similarly, the Volga is the sensitivity of the Vega with respect to a change of the implied volatility :
Volga =
V
I = X
BS
+X
vega
vega
+X
vanna
vanna
+X
volga
volga
(19)
9
where
I =
_
_
0
RR
mkt
RR
BS
BF
mkt
BF
BS
_
_
_
_
vega
vanna
volga
_
_
= (A
T
)
1
I (20)
and where the quantities
i
can be interpreted as the market prices attached to a unit amount of
Vega, Vanna and Volga, respectively. For vanillas this gives a very good approximation of the market
price. For exotics, however, e.g. no-touch options close to a barrier, the resulting correction typically
turns out to be too large. Following market practice we thus modify (19) to
X
VV
= X
BS
+p
vanna
X
vanna
vanna
+p
volga
X
volga
volga
(21)
where we have dropped the Vega contribution which turns out to be several orders of magnitude smaller
than the Vanna and Volga terms in all practical situations, and where p
vanna
and p
volga
represent
attenuation factors which are functions of either the survival probability or the expected rst-exit
time. We will return to these concepts in section 5.
4.2 Vanna-Volga as a smile-interpolation method
In [5], Castagna and Mercurio show how Vanna-Volga can be used as a smile interpolation method.
They give an elegant closed-form solution (unique) of system (17), when X is a European call or put
with strike K.
In their paper they adjust the Black Scholes price by using a replicating portfolio composed of a
weighted sum of three vanillas (calls or puts) struck respectively at K
1
, K
2
and K
3
, where K
1
< K
2
<
K
3
. They show that the weights w
i
associated to the vanillas struck at K
i
such that the resulting
portfolio hedges the Vega, Vanna and Volga risks of the vanilla of strike K are unique and given by:
w
1
(K) =
Vega(K)
Vega(K
1
)
ln
K
2
K
ln
K
3
K
ln
K
2
K
1
ln
K
3
K
1
w
2
(K) =
Vega(K)
Vega(K
2
)
ln
K
K
1
ln
K
3
K
ln
K
2
K
1
ln
K
3
K
2
(22)
w
3
(K) =
Vega(K)
Vega(K
3
)
ln
K
K
1
ln
K
K
2
ln
K
3
K
1
ln
K
3
K
2
The fact that this solution provides an exact interpolation method is easily veried by noticing that
w
i
(K
i
) = 1 and w
i
(K
j
) = 0, i ,= j.
This solution still holds in the case of a replicating portfolio composed of ATM, RR and BF
instruments as described in section 4 by equations (14). Setting K
1
= K
p
, K
2
= K
ATM
and K
3
= K
c
,
a simple coordinate transform yields:
w
ATM
(K) = w
1
(K) +w
2
(K) +w
3
(K)
w
RR
(K) =
1
2
(w
3
(K) w
1
(K)) (23)
w
BF
(K) = w
1
(K) +w
3
(K)
where the weights w
ATM
, w
RR
and w
BF
are dened by (17)-(18).
10
We now turn back to the elementary Vanna-Volga recipe (15). Unlike the previously exposed exact
solution, it does not reproduce the market price of RR and BF, a fortiori is it not an interpolation
method for plain vanillas. However, this approximation possesses the merit of allowing a qualitative
interpretation of the RR and BF correction terms in (15).
As we will demonstrate, those two terms directly relate to the slope and convexity of the smile
curve. To start with, we introduce a new smile parametrization variable:
Y = ln
K
F exp (
1
2
2
ATM
)
= ln
K
K
ATM
Note that the Vega of a Plain Vanilla Option is a symmetric function of Y :
Vega(Y ) = Vega(Y ) = S e
r
f
n
_
Y
_
where n() denotes the Normal density function.
Let us further assume that the smile curve is a quadratic function of Y :
(Y ) =
ATM
+bY +cY
2
(24)
In this way we allow the smile to have a skew (linear term) and a curvature (quadratic term), while
keeping an analytically tractable expression. We now express Vanna and Volga of Plain Vanilla Options
as functions of Y :
Vanna(Y ) = Vega(Y )
Y +
2
S
2
Volga(Y ) = Vega(Y )
Y
2
+
2
Y
(25)
Working with the plain Black-Scholes Delta (2) and the delta-neutral ATM denition and dening
Y
i
= ln
K
i
K
ATM
we have that Y
ATM
and Y
25P
and Y
25C
corresponding respectively to At-The-Money,
25-Delta Put, and 25-Delta Call solve
Y
ATM
= 0
DF
f
(t, T) N
_
Y
25P
1
2
(
2
25P
2
ATM
)
25P
_
=
1
4
DF
f
(t, T) N
_
Y
25C
+
1
2
(
2
25C
2
ATM
)
25C
_
=
1
4
Under the assumption that
25C
25P
ATM
we nd Y
25C
Y
25P
. In this case using
equations (25) and (14), the Vanna of the RR and the Volga of the BF can be expressed as :
Vanna(RR) = Vanna(Y
25C
) Vanna(Y
25P
) = 2
Vega(Y
25C
)Y
25C
S
2
ATM
Volga(BF) =
Volga(Y
25C
) + Volga(Y
25P
)
2
Volga(0) =
Vega(Y
25C
)Y
2
25C
3
ATM
(26)
11
To calculate RR
cost
and BF
cost
(the dierence between the price calculated with smile, and that
calculated with a constant volatility
ATM
), we introduce the following convenient approximation:
Call((Y )) Call(
ATM
) Vega(Y ) ((Y )
ATM
) (27)
using the above, it is straightforward to show that:
RR
cost
2b Vega(Y
25C
)Y
25C
BF
cost
cVega(Y
25C
) Y
2
25C
(28)
Substituting expressions (25), (26) and (28) in the simple VV recipe (15) yields the following remark-
ably simple result:
X
VV
(Y ) = X
BS
(Y ) +
Vanna(Y )
Vanna(RR)
RR
cost
+
Volga(Y )
Volga(BF)
BF
cost
X
BS
+ Vega(Y)bY + Vega(Y)cY
2
+ Vega(Y)
2
ATM
(b +cY ) (29)
= X
BS
(Y ) +
X
BS
(Y ) ((Y )
ATM
) + Vega(Y)
2
ATM
(b +cY )
Despite the presence of a residual term, which vanishes as 0 or
ATM
0, the above ex-
pression shows that the Vanna-Volga price (15) of a vanilla option can be written as a rst-order
Taylor expansion of the BS price around
ATM
. Furthermore, as
Vanna(Y )
Vanna(RR)
RR
cost
Vega(Y)bY and
Volga(Y )
Volga(BF)
BF
cost
Vega(Y)cY
2
, the RR term (coupled to Vanna) accounts for the impact of the linear
component of the smile on the price, while the BF (coupled to Volga) accounts for the impact of the
quadratic component of the smile on the price.
5 Market-adapted variations of Vanna-Volga
In this section we describe two empirical ways of adjusting the weights
_
p
vanna
, p
volga
_
in (21). We
will focus our attention on knock-out options, although the Vanna-Volga approach can be readily
generalized to options containing knock-in barriers, as those can always be decomposed into two
knock-out (or vanilla) ones (through the no-arbitrage relation knock-in = vanilla knock-out).
To justify the need for the correction factors to (21) we argue as follows: As the knock-out barrier
level B of an option is gradually moved toward the spot level S
t
, the BS price of a KO option must
be a monotonically decreasing function, converging to zero exactly at B = S
t
. Since the Vanna-Volga
method is a simple rule-of-thumb and not a rigorous model, there is no guarantee that this will be
satised. We thus have to impose it through the attenuations factors p
vanna
and p
volga
. Note that for
barrier values close to the spot, the Vanna and the Volga behave dierently: the Vanna becomes large
while, on the contrary, the Volga becomes small. Hence we seek attenuation factors of the form:
p
vanna
= a p
volga
= b +c (30)
where [0, 1] represents some measure of the barrier(s) vicinity to the spot with the features
= 0 for S
t
B (31)
= 1 for [S
t
B[ 0 (32)
12
i.e. the limiting cases refer to the regions where the spot is close versus away from the barrier level.
Before moving to more specic denitions of , let us introduce some restrictions on p
vanna
and p
volga
:
lim
1
p
vanna
= 1 lim
1
p
volga
= 1 (33)
The above conditions ensure that when the barrier is far from the spot, implying that hitting the
barrier becomes very unlikely, the Vanna-Volga algorithm boils down to its simplest form (15) which
is a good approximation to the price of a vanilla option using the market quoted volatility. We
therefore amend the expressions (30) into continuous, piecewise linear functions:
p
vanna
=
_
a
a
1
1
>
p
volga
=
_
b +c
(b +c
)
1
1
>
(34)
where
is a transition threshold chosen close to 1. Note that the amendment (34) is justied only
in the case of options that degenerate into plain vanilla instruments in the region where the barriers
are away from the spot. However, in the case of treasury options that do not have a strike (e.g. OT),
there is no smile eect in the region where the barriers are away from the spot as these options pay a
xed amount and their fair value is provided by the BS TV. In this case, no amendment is necessary
as both Vanna and Volga go to zero.
We now proceed to specify practical candidates, namely the survival probability and the expected
rst exit time (FET). In what follows, the corresponding Vanna-Volga prices will be denoted by VV
surv
and VV
fet
respectively.
5.1 Survival probability
The survival probability p
surv
[0, 1] refers to the probability that the spot does not touch one or
more barrier levels before the expiry of the option. Here we need to distinguish whether the spot
process is simulated through the domestic or the foreign risk-neutral measures:
domestic : dS
t
= S
t
(r
d
r
f
) dt +S
t
dW
t
(35)
foreign : dS
t
= S
t
(r
d
r
f
+
2
) dt +S
t
dW
t
(36)
where W
t
is a Wiener process. One notices that the quanto drift adjustment will obviously have an
impact in the value of the survival probability. Then, for e.g. a single barrier option we have
domestic : p
d
surv
= E
d
[1
S
t
<B,t<t
<T
] = NT
d
(B)/DF
d
(t, T) (37)
foreign : p
f
surv
= E
f
[1
S
t
<B,t<t
<T
] = NT
f
(B)/DF
f
(t, T) (38)
where NT
d/f
(B) is the value of a no-touch option in the domestic/foreign measure, E
d/f
is the risk
neutral expectation in the domestic/foreign market respectively, and 1
a
is the indicator function for
the event a. Similarly, for options with two barriers the survival probability is given through the
undiscounted value of a double-no-touch option. Explicit formulas for no-touch and double-no-touch
options can be found in [25].
The survival probability clearly satises the required features (31), (32). To respect domes-
tic/foreign symmetries we further dene
surv
=
1
2
(p
d
surv
+p
f
surv
).
13
5.2 First exit time
The rst exit time is the minimum between: (i) the time in the future when the spot is expected
to exit a barrier zone before maturity, and (ii) maturity, if the spot has not hit any of the barrier
levels up to maturity. That is, if we denote the FET by u(S
t
, t) then u(S
t
, t) = min, where
= inf [0, )[S
t+
> H or S
t+
< L where L < S
t
< H dene the barrier levels, S
t
the spot of
today and the time to maturity (expressed in years). This quantity also has the desirable feature
that it becomes small near a barrier and can therefore be used to rescale the two correction terms in
(21).
Let us give some denitions. For a geometric Brownian motion spot process of constant volatility
and drift , the cumulative probability of the spot hitting a barrier between t
and t
(t < t
< T,
t
> t
) denoted by C(S, t
, t
, t
) can be
thought of as the undiscounted price of a DOT option):
T C = 0 T
t
+
1
2
2
S
2
2
S
2
+S
S
(39)
with boundary conditions C(L, t
, t
) = C(H, t
, t
) = 1 and C(S, t
, t
) = t
t +
_
T
t
(t
)
C
t
dt
+
_
T
(T t
)
C
t
dt
(40)
while integration by parts gives
u(S, t
) = t
t +
_
T
t
_
1 C(S, t
, t
)
_
dt
(41)
and nally taking derivative with respect to t
to T results in:
u
t
+
1
2
2
S
2
2
u
S
2
+S
u
S
= 0 T u = 0 (42)
note that this is slightly dierent from the expression in [24], where FET is measured from t
. Equation
(42) is solved backwards in time from t
= T to t
) = u(H, t
) = t
. Note that
we have divided by the time to maturity in order to have a dimensionless quantity with
fet
[0, 1].
2
In a window-barrier option, the barrier is activated at a time greater than the selling time of the option and deactivates
before the maturity of the option.
14
Figure 2: Comparison between
surv
and
fet
plotted against barrier level, in a single barrier case (left
panel) and a double barrier case (right panel). Used Market Data: S=1.3, =1.3, r
d
=5%, r
f
=3%
ATM
=20%
5.3 Qualitative dierences between
surv
and
fet
Although
surv
and
fet
possess similar asymptotic behavior (converging to 0 for options innitely
close to knocking-out, converging to 1 for an option innitely far from knocking-out), they represent
dierent quantities, and can dier substantially in intermediate situations. To support this assertion,
we show in Figure 2 plots of
surv
and
fet
as a function of the barrier level, in a single-barrier and in
a double-barrier case. While in the single barrier case the shapes of the two curves look similar, their
discrepancy is more pronounced in the double-barrier case where the upper barrier is kept constant,
and the lower barrier is progressively moved away from the spot level. For barrier levels close to the
spot, there is a plateau eect in the case of
surv
which stays at zero, while
fet
seems to increase
linearly. This can be explained intuitively: moving the barrier level in the close vicinity of the spot
will not prevent the spot from knocking out at some point before maturity (hence
surv
0). But
although the knocking event is almost certain, the expected time at which it occurs directly depends
on the barrier-spot distance.
This discussion should emphasize the importance of a careful choice between the two candidates,
especially when it comes to pricing double-barrier options.
There is no agreed consensus regarding which of
surv
,
fet
is a better candidate for in (30). Based
on empirical observations, it is suggested in e.g. [23] that one uses
surv
with a = 1 and b = c = 0.5.
Other market beliefs however favor using
fet
with a = c = 1 and b = 0. In [26], the absence
of mathematical justication for these choices is highlighted, and other adjustment possibilities are
suggested, depending on the type of option at hand. In section 6 we will discuss a more systematic
procedure that can allow one to calibrate the Vanna-Volga model and draw some conclusions regarding
the choice of pricer.
15
5.4 Arbitrage tests
As the Vanna-Volga method is not built on a solid bedrock but is only a practical rule-of-thumb, there
is no guarantee that it will be arbitrage free. Therefore as part of the pricer one should implement a
testing procedure that ensures a few basic no-arbitrage rules for barrier options (with or without strike):
For example, (i) the value of a vanilla option must not be negative, (ii) the value of a single/double
knock-out barrier option must not be greater than the value of the corresponding vanilla, (iii) the
value of a double-knock-out barrier option must not be more expensive than either of the values of the
corresponding single knock-outs, (iv) the value of a window single/double knock-out barrier option
must be smaller than that of the corresponding vanilla and greater than the corresponding american
single/double knock-out. For knock-in options, the corresponding no-arbitrage tests can be derived
from the replication relations: (a) for single barriers, KI(B) = VAN KO(B), where B represents the
barrier of the option, and (b) for double-barriers, KIKO(KIB,KOB) = KO(KIB) DKO(KIB,KOB)
where KIB and KOB represent the knock-in and knock-out barrier respectively.
For touch or no-touch options, the above no-arbitrage principles are similar. One-touch options
can be decomposed into a discounted cash amount and no-touch options: OT(B) = DF - NT(B) and
similarly for double-one-touch options.
Based on these principles a testing procedure can be devised that amends possible arbitrage in-
consistencies. We begin by using replication relations to decompose the option into its constituent
parts if needed. This leaves us with vanillas and knock-out options for which we calculate the BSTV
and the Vanna-Volga correction. On the resulting prices we then impose
VAN = max(VAN, 0) KO = max(KO, 0) (45)
to ensure condition (i) above. We then proceed with imposing conditions (ii)-(iv):
KO = min(KO, VAN) WKO = min(WKO, VAN) WKO = max(WKO, KO) (46)
while for double-knock-out options we have in addition
DKO = min(DKO, KO(1)) DKO = min(DKO, KO(2)) (47)
where KO(1) and KO(2) represent the corresponding single knock-out options.
Note that both in the case of a double-knock-out and in that of a window-knock-out we need to
create a single-knock-out instrument and launch a no-arbitrage testing on it as well.
As an example, let us consider a window knock-in knock-out option. Having an in barrier this
option will be decomposed to a dierence between a window knock-out and a window double knock-
out. For the former, we will create the corresponding KO option while for the latter the corresponding
DKO. In addition, we will also need the plain vanilla instrument. We will then price the KO and
DKO separately using the Vanna-Volga pricer, ensure that the resulting value of each of these is
positive (equation (45)), impose condition (iii) (equation (47)) to ensure no-arbitrage on the DKO
and condition (iv) (equation (46)) to ensure that the barrier options are not more expensive than the
plain vanilla.
5.5 Sensitivity to market data
As the FX derivatives market is rife with complex conventions it can be the case that pricing errors
stemming from wrong input data have a greater impact than errors stemming from assuming wrong
16
smile dynamics. This warrants discussion concerning the sensitivity of FX models with respect to
market data. Already from (15) we can anticipate that the Vanna-Volga price is sensitive to the
values of
RR25
and
BF25(2vol)
. To emphasize this dependency we will consider the following two
sensitivities:
RR
=
d Price
d
RR25
BF
=
d Price
d
BF25(2vol)
(48)
which measure the change in the Vanna-Volga price given a change in the input market data. In our
tests we have used the Vanna-Volga survival probability for a series of barrier levels of a OT option.
Similar considerations follow by using the FET variant. The results are shown in Figure 3 where on
top of the two sensitivities we superimposed the Vanna and the Volga of the option.
We notice that the two sensitivities can deviate signicantly away from zero. This highlights the
importance of using accurate and well-interpreted market quotes. For instance, in the 1-year USDCHF
OT with the touch-level at 1.55 (BSTV price is 4%), an error of 0.5% in the value of
BF25(2vol)
would induce a price shift of 3%. This is all but negligible! Thus a careful adjustment of the market
data quotes is sometimes as important as the model selection.
We also see that the Volga provides an excellent estimate of the models sensitivity to a change in
the Buttery values. Similarly, Vanna provides a good estimate of the models sensitivity to a change
in the Risk Reversal values but only as long as the barrier level is suciently away from the spot.
This disagreement in the region close to the spot is linked to the fact that in the Vanna-Volga recipe of
section 5.1 we adjusted the Vanna contribution by the survival probability which becomes very small
close to the barrier.
Figure 3 implies that for all practical purposes one should be on guard for high BS values of Vanna
and/or Volga which indicate that the pricer is sensitively dependent on the accuracy of the market
data.
6 Numerical results
In order to assess the ability of the Vanna-Volga family of models (21) to provide market prices, we
compared them to a large collection of market indicative quotes. By indicative we mean that the
prices we collected come from trading platforms of three major FX-option market-makers, queried
without eectively proceeding to an actual trade. It is likely that the models behind these prices do
not necessarily follow demand-supply dynamics and that the providers use an analytic pricing method
similar to the Vanna-Volga we present here.
Our pool of market prices comprises of 3-month and 1-year options in USDCHF and USDJPY, the
former currency pair typically characterized by small
RR
values, while the latter by large ones. In this
way we expect to span a broad range of market conditions. For each of the four maturity/currency
pair combinations we select four instrument types, representative of the rst generation exotics family:
Reverse-Knock-Out call (RKO), One-Touch (OT), Double-Knock-Out call (DKO), and Double-One-
Touch (DOT). In the case of single barrier options (RKO and OT), 8 barrier levels are adjusted,
mapping to probabilities of touching the barrier that range from 10% to 90%. In the case of the RKO
call, the strike is set At-The-Money-Spot. In the case of two-barrier options (DKO and DOT), since it
is practically impossible to fully span the space of the two barriers we selected the following subspace:
(i) we x the lower barrier level in such a way that it has a 10% chance of being hit, then select 5 upper
barrier levels such that the overall hitting probabilities (of any of the 2 barriers) range approximately
from 15% to 85%. (ii) We repeat the same procedure with a xed upper barrier level, and 5 adjusted
lower barrier levels.
17
Figure 3: Sensitivity of the Vanna-Volga price with respect to input market data for a OT option. Top:
Comparison between the Vanna (BSTV) and
RR
. Bottom: Comparison between the Volga (BSTV)
and
BF
. We see that the two Greeks provide a good approximation of the two model sensitivities.
18
In summary, our set of data consists of the cross product T of the sets
currency pair : A = USDJPY, USDCHF
maturity period : B = 3m, 1y
option type : C = RKO, OT, DKO, DOT
barrier value : D = B
1
, . . . , B
n
(49)
where n = 10 for double-barrier options and n = 8 for single barrier ones.
In order to maintain coherence, each of the two data sets were collected in a half-day period (in
Nov. 2008 for USDJPY, in Jan. 2009 for USDCHF).
Thus in total our experiments are run over the set of models
models : E = VV
surv
, VV
fet
(50)
6.1 Denition of the model error
In order to focus on the smile-related part of the price of an exotic option, let us dene for each
instrument i T from our pool of data (49) the Model Smile Value (MODSV) and the Market
Smile Value (MKTSV) as the dierence between the price and its Black-Scholes Theoretical Value
(BSTV):
MODSV
k
i
= Model Price
k
i
BSTV
k
i
k = 1, . . . , N
mod
MKTSV
k
i
= Market Price
k
i
BSTV
k
i
k = 1, . . . , N
mkt
(where market prices are taken as the average between bid and ask prices) and where N
mod
= 4 is
the number of models we are using and N
mkt
= 3 the number of FX market makers where the data is
collected from. Let us also dene the average, minimum and maximum of the market smile value:
MKTSV
i
=
1
N
mkt
kN
mkt
MKTSV
k
i
min
i
= min
kN
mkt
MKTSV
k
i
, max
i
= max
kN
mkt
MKTSV
k
i
(51)
We now introduce an error measure quantifying the ability of a model to describe market prices. This
function is dened as a quadratic sum over the pricing error :
k
=
iF
_
MODSV
k
i
MKTSV
i
max
i
min
i
_
2
(52)
The error is weighted by the inverse of the market spread, dened as the dierence between the
maximum and the minimum mid market price for a given instrument. This setup is designed (i)
to yield a dimensionless error measure that can be compared across currency pairs and the type of
options, (ii) to link the error penalty to the market coherence: a pricing error on an instrument which
is priced very similarly by the 3 market providers will be penalized more heavily than the same pricing
error where market participants exhibit large pricing dierences among themselves. Note also that
the error is dened as the deviation from the average market price.
19
6.2 Shortcomings of common stochastic models in pricing exotic options
Before trying to calibrate the Vanna-Volga weighting factors p
vanna
and p
volga
, we investigate how the
Dupire local vol [6] and the Heston stochastic vol [7] models perform in pricing our set of selected exotic
instruments (for a discussion on the pricing of barrier instruments under various model frameworks,
see for example [13, 14, 15]). In order to obtain a fast and reliable calibration for Heston, the price of
call options is numerically computed through the characteristic function [1, 11], and Fourier inversion
methods. To price exotic options, Heston dynamics is simulated by Monte Carlo, using a Quadratic-
Exponential discretization scheme [2].
Figure 4 shows the MODSV of a 1-year OT options in USDCHF (lower panel) and USDJPY (upper
panel), as the barrier moves away from the spot level (S
t
= 95.47 for USDJPY and S
t
= 1.0902 for
USDCHF). At rst inspection, none of the models gives satisfactory results.
USDCHF USDJPY
RKO Heston Dupire
OT Heston Heston
DKO(Up) Dupire Dupire
1-Year DKO(Down) Heston Heston
DOT(Up) Heston Dupire
DOT(Down) Heston Dupire
global Heston ( = 62) Dupire ( = 96)
RKO Heston Dupire
OT Heston Dupire
DKO(Up) Heston Dupire
3-Month DKO(Down) Heston Heston
DOT(Up) Heston Dupire
DOT(Down) Heston Heston
global Heston ( = 65) Dupire ( = 73)
Table 2: Heston stochastic vol Vs. Dupire local vol in pricing 1st generation exotics.
Using the error measure dened above, we now try to formalize the impressions given by our rough
inspection of Figure 4. For each combination of the instruments in (49) we determine which of Dupire
local vol or Heston stochastic vol gives better market prices. The outcome of this comparison is given
in the Table 2.
This table suggests that in a simplied world where exotic option prices derive either from Dupire
local vol or from heston stochastic vol dynamics an FX market characterized by a mild skew (US-
DCHF) exhibits mainly a stochastic volatility behavior, and that FX markets characterized by a
dominantly skewed implied volatility (USDJPY) exhibit a stronger local volatility component. This
conrms that calibrating a stochastic model to the vanilla market is by no mean a guarantee that
exotic options will be priced correctly [21], as the vanilla market carries no information about the
smile dynamics.
In reality the market dynamics could be better approximated by a hybrid volatility model that
contains both some stochastic vol dynamics and some local vol one. This model will be quite rich but
the calibration can be expected to be considerably hard, given that it tries to mix two very dierent
smile dynamics, namely an absolute local-vol one with a relative stochastic vol one. For a discussion
20
Figure 4: Smile value vs. barrier level; comparison of the various models for OT 1-year options in
USDCHF (bottom) and USDJPY (top). Market limits are indicated with black solid lines.
21
of such a model, we refer the reader to [14].
At this stage one has the option to either go for the complex hybrid model or for the more heuristic
alternative method like the Vanna-Volga. In this paper we present the latter.
6.3 Vanna-Volga calibration
The purpose of this section is to provide a more systematic approach in selecting the coecients a, b
and c in (30) and thus the factors p
vanna
and p
volga
.
We rst determine the optimal values of coecients a, b and c in the sense of the least error (52),
where the sum extends to all instruments and to the two maturities (e.g. a single error function per
currency pair). This problem can readily be solved using standard linear regression tools, as a, b and
c appear linearly in the VV correction term, but most standard solver algorithms would as well do the
job. This optimization problem is solved four times in total, for USDCHF with
surv
and
fet
, and for
USDJPY with
surv
and
fet
. Let us point out that such a calibration is of course out of the question
in a real trading environment: collecting such an amount of market data each time a recalibration
is deemed necessary would be way too time-consuming. Our purpose is simply to determine some
limiting cases, to be used as benchmarks for the results of a more practical calibration procedure
discussed later. Table 3 presents these optimal solutions, indicating the minimum error value, along
with the value of the optimal coecients a, b and c.
USDCHF USDJPY
surv
= 19.7 = 15.6
a = 0.54, b = 0.29, c = 0.14 a = 0.74, b = 0.7, c = 0.05
fet
= 18.2 = 14.7
a = 0.49, b = 0.35, c = 0.01 a = 0.54, b = 0.17, c = 0.52
Table 3: Overall pricing error, calibration on entire market price set.
Comparing the above error numbers to those of Table 2, it seems possible that the Vanna-Volga
models have the potential to outperform the Dupire or Heston models.
We now discuss a more practical calibration approach, where the minimization is performed on OT
prices only. The question we try to answer is: Can we calibrate a VV model on OT market prices,
and use this model to price other rst generation exotic products ?. Performing this calibration
with 3 parameters to optimize will certainly improve the tting of OT prices, but at the expense of
destroying the tting quality for the other instruments (in the same way that performing a high-order
linear regression on a set of data points, will produce a perfect match on the data points and large
oscillations elsewhere). This is conrmed by the results of Table 4, showing how the error (on the
entire instrument set) increases with respect to the error of Table 3 when the optimization is performed
on the OT subset only.
USDCHF USDJPY
surv
= 44.6 = 26.8
fet
= 47.2 = 85
Table 4: Overall pricing error, calibration on OT prices only.
22
For robustness reasons, it is thus desirable to reduce the space of free parameters in the optimization
process. We consider the following two constrained optimization setups: (i) a = c, b = 0 and (ii)
b = c = 0.5 a, which are re-scaled versions of the market practices described in section 5.3. Needless
to say that the number of possible congurations here are limited only by ones imagination. Our
choice is dictated mainly by simplicity, namely we have chosen to keep a single degree of freedom.
The results are presented in Table 5 where we compare four possible congurations measured over all
instruments and maturity periods for our two currency pairs.
USDCHF USDJPY Total error
conguration 1
surv
= 21.8 = 28.4 =50.2
b = c = 0.5 a a = 0.43 a = 0.72
conguration 2
fet
= 21.2 = 26 =47.2
b = c = 0.5 a a = 0.39 a = 0.63
conguration 3
surv
= 32.2 = 72.1 =104.3
a = c, b = 0 a = 0.51 a = 0.69
conguration 4
fet
= 24.3 = 19.4 =43.7
a = c, b = 0 a = 0.42 a = 0.60
Table 5: Overall pricing error, constrained calibration on OT prices only.
As there is no sound mathematical (or economical) argument to prefer one conguration over
another, we therefore choose the least-error conguration, namely conguration n
o
4. One additional
argument in favor of
fet
is that it accommodates window-barrier options without further adjustment.
This is not the case of
surv
where some re-scaling should be used to account for the start date of the
barrier (when the barrier start date is very close to the option maturity, the path-dependent character
vanishes and the full VV correction applies i.e. p
vanna
= p
volga
= 1 even for small
surv
values).
In Figure 5 we show results from the calibration of the Vanna-Volga method. It is based on
minimizing the error (52) of (i) all instruments of the data pool and while having all coecients a, b, c
of
fet
free and (ii) of one-touch options only and with conguration n
o
4 (thus, we have chosen
fet
with a = c, b = 0). We see that in general calibration (i) performs better in the sense that it falls
well within the shaded area that corresponds to the limits of the market price as provided by the FX
market makers. This is not surprising as this calibration is meant to be the most general and exible.
However this is clearly an impractical calibration procedure. On the contrary, the calibration method
(ii) that is based on quotes from a single exotic instrument has practical advantages and appears in
good agreement with that of (i). Finally note that these pictures are representative of our results in
general.
7 Conclusion
The Vanna-Volga method is a popular pricing tool for FX exotic options. It is appealing to both
traders, due to its clear interpretation as a hedging tool, and to quantitative analysts, due to its
simplicity, ease of implementation and computational eciency. In its simplest form, the Vanna-Volga
recipe assumes that smile eects can be incorporated to the price of an exotic option by inspecting the
eect of the smile on vanilla options. Although this recipe, outlined in (15), turns out to give often
uncomfortably large values, there certainly is a silver lining there. This has led market practitioners
23
Figure 5: Results from calibrating the Vanna-Volga method on (i) all instruments of our data pool (marked as
VV opt (global)), (ii) one-touch options only (marked as VV opt (OT)). The results of the two calibrations
do not dier signicantly while the latter is naturally more convenient from a practical perspective. The shaded
areas correspond to the region within which market makers provide their indicative mid price. For comparison
we also show the non-calibrated Vanna-Volga methods based on the survival probability and the rst exit
time.
24
to consider several ways to adapt the Vanna-Volga method. In this article we have reviewed some
commonly used adaptations based on rescaling the Vanna-Volga correction by a function of either the
survival probability or the rst exit time. These variations provide prices that are more in line with
the indicative ones given by market makers.
We have attempted to improve the Vanna-Volga method further by adjusting the various rescaling
factors that are involved. This optimization is based on simple data analysis of one-touch options that
are obtained from renowned FX platforms. It involves a single optimization variable and as a result
we nd that for a wide range of exotic options, maturity periods and currency pairs it leads to prices
that agree well with the market mid-price.
The FX derivatives community, perhaps more than any other asset class, lives on a complex
structure of quote conventions. Naturally, a wrong interpretation of the input market data cannot
lead to the correct results. To this end, we have presented some relevant FX conventions regarding
smile quotes and we have tested the robustness of the Vanna-Volga method against the input data. It
appears that the values of Vanna and Volga provide a good indication of the VV price sensitivity to
a change in smile input parameters.
References
[1] H. Albrecher, P. Mayer, W. Schoutens and J. Tistaert, The little Heston trap,Wilmott Magazine
(January 2007) 83-92.
[2] L. Andersen, Ecient Simulation of the Heston Stochastic Volatility Model, Available at SSRN:
http://ssrn.com/abstract=946405 (January 23, 2007).
[3] F. Black and M. Scholes, The pricing of Options and Corporate Liabilities, Journal of Political
Economy 81 (1973) 637-654.
[4] D. Brigo and F. Mercurio, Interest Rate Models Theory and Practice With Smile, Ination
and Credit, Springer (2007).
[5] A. Castagna and F. Mercurio, The Vanna-Volga method for implied volatilities, Risk (January
2007) 106-111.
[6] B. Dupire, Pricing with a smile, Risk (January 1994) 18-20.
[7] S. L. Heston, A closed-form solution for options with stochastic volatility with applications to
bond and currency options, Rev Fin Studies 6 (1993) 327-343.
[8] J. C. Hull, Options, Futures and Other Derivatives, Prentice Hall Series in Finance (2006) 6th
edition.
[9] E. Reiner and M. Rubinstein, Unscrambling the Binary Code, Risk October 1991, 75-83.
[10] E. Reiner and M. Rubinstein, Breaking down the barriers, Risk September 1991, 28-35.
[11] P. Jackel and C. Kahl, Not-So-Complex Logarithms in the Heston Model, Wilmott Magazine
(September 2005) 94-103.
[12] N. Kunitomo and M. Ikeda, Pricing Options with Curved Boundaries, Mathematical Finance 4
(1992) 275-298.
25
[13] A. Lipton, Mathematical Methods for Foreign Exchange: A Financial Engineers Approach,
World Scientic (2001).
[14] A. Lipton, The vol smile problem, Risk Magazine (2002) 15, n. 2, 61-65.
[15] A. Lipton and W. McGhee, Universal barriers, Risk Magazine (2002) 15, n. 5, 81-85.
[16] T. Fisher, Variations on the Vanna-Volga adjustment, Bloomberg Quantitative Research and
Development FX Team (2007), private communication.
[17] Y. Shkolnikov, Generalized Vanna-Volga Method and its Applications, Available at SSRN:
http://ssrn.com/abstract=1186383 (June 25, 2009).
[18] R. C. Merton, Theory of Rational Option Pricing, Bell Journal of Economics and Management
Science 4 (1973) 141-183.
[19] E. Derman, Regimes of Volatility, Quantitative Strategies Research Notes, Goldman Sachs,
(1999).
[20] M. Rubinstein and E. Reiner, Exotic Options, Working paper, UC Berkeley (1992).
[21] W. Schoutens, E. Simons, and J. Tistaert, A Perfect calibration! Now what?, Wilmott Magazine
(March 2004).
[22] S. E. Shreve, Stochastic Calculus for Finance II: Continuous-Time Models, Springer-Finance
(2004).
[23] H. J. Stein, FX Market Behavior and Valuation, Available at SSRN:
http://ssrn.com/abstract=955831 (December 13, 2006).
[24] P. Wilmott, Paul Wilmott on Quantitative Finance, John Wiley & Sons (2006).
[25] U. Wystup, FX Options and structured products, Wiley Finance (2006).
[26] U. Wystup, Vanna-Volga Pricing, MathFinance AG (June 2008) Available at
http://www.mathnance.com/wystup/papers/wystup
vannavolga
eqf.pdf.
[27] U. Wystup, The market price of one-touch options in foreign exchange markets, Derivatives
Week (2003) 12(13), 1-4.
26
A Denitions of notation used
t date of today
T maturity date
= (T t)/365 time to expiry (expressed in years)
S
t
spot today
K strike
r
f/d
(t) foreign/domestic interest rates
volatility of the FX-spot
DF
f/d
(t, T) = exp[r
f/d
] foreign/domestic discount factor
F = S
t
DF
f
(t, T)/DF
d
(t, T) forward price
d
1
=
ln
F
K
+
1
2
d
2
=
ln
F
K
1
2
N(z) =
_
z
dx
1
2
e
1
2
x
2
cumulative normal
Table 6: List of abbreviations.
B Premium-included Delta
For correctly calculating the Delta of an option it is important to identify which of the currencies
represents the risky asset and which one represents the riskless payment currency.
Let us consider a generic spot quotation in terms Ccy1-Ccy2 representing the amount of Ccy2
per unit of Ccy1. If the (conventional) premium currency is Ccy2 (e.g. USD in EURUSD) then by
convention the risky asset is Ccy1 (EUR in this case) while Ccy2 refers to the risk-free one. In this
case the standard Black-Scholes theory applies and the Delta expressed in Ccy1 is found by a simple
dierentiation of (1):
BS
= DF
f
(t, T) N(d
1
). This represents an amount of Ccy1 to sell if one is long
a Call.
If, however, the premium currency is Ccy1 (e.g. USD in USDJPY) then Ccy2 is considered as the
risky asset while Ccy1 the risk-free one. In this case, the value of the Delta is = S
t
BS
Call
t
, where
Call
t
is the premium in units of Ccy2 while and
BS
are expressed in their natural currencies;
Ccy2 and Ccy1, respectively (for lightening notations, we omit the time index t in and
BS
). In
this case represents an amount of Ccy2 to buy. This relation can be seen by the following argument.
First note that the Black-Scholes vanilla price of a call option is
Call
t
= DF
d
(t, T) E
d
_
max(S
T
K, 0)
(53)
where the index d implies that we are referring to the domestic risk-neutral measure, i.e. we take the
domestic money-market (MM) unit 1/DF
d
(t, T) as numeraire. If we now wish to express (53) into a
measure where the numeraire is the foreign money-market account then
Call
t
= DF
d
(t, T) E
d
_
max(S
T
K, 0)
= DF
d
(t, T) E
f
_
dQ
d
dQ
f
(T) max(S
T
K, 0)
(54)
27
where we introduced the Radon-Nikodym derivative (see for example [4, 22])
dQ
d
dQ
f
(T) =
DF
f
(t, T)
DF
d
(t, T)
S
t
S
T
(55)
This equality allows us to derive the foreign-domestic parity relation
Call
t
= DF
d
(t, T) E
d
_
max(S
T
K, 0)
= DF
f
(t, T) S
t
KE
f
_
max(
1
K
1
S
T
, 0)
_
(56)
where both sides are expressed in units of Ccy2 (for a unit nominal amount in Ccy1). The above
foreign/domestic relation illustrates the fact that in FX any derivative contract can be regarded either
from a domestic or from a foreign standpoint. However the contract value is unique. On the contrary,
the Delta of the option depends on the adopted perspective. In domestic vs. foreign worlds we have
respectively
BS
=
Call
t
S
t
=
Callt
St
1
St
(57)
where the rst equation is expressed in units of Ccy1 (to sell) while the second in units of Ccy2 (to
buy). Setting up a Delta hedged portfolio (at time t) in the foreign world implies that at any instant
of time t
t
=
Call
t
S
t
+
S
t
(58)
will be insensitive to variations of the spot S
t
. From
t
/S
t
[
t
=t
= 0 we then nd
= S
t
BS
Call
t
(59)
Note that FX convention dictates that the is always quoted in units of Ccy1 (regardless of the
currency to which the premium is paid), hence to obtain the relation mentioned in section 3.1 we
simply take
= /S
t
. Table 7 provides a vis-`a-vis of the various quantities under the two
perspectives for an option in USDJPY with the Spot S
t
dened as the amount of JPY per USD.
USD world JPY world
Local MM unit 1 USD 1 JPY
Risky asset JPY USD
Contract value in local MM units Call
t
/S
t
Call
t
Risky asset in local MM units 1/S
t
S
t
hedge: amount of risky asset to short
Call
t
S
t
1
S
t
= Call
t
S
t
BS
(JPY)
Call
St
=
BS
(USD)
Amount of USD to short
=
Call
S
t
1
S
t
1
St
=
BS
1
St
Call (USD)
Call
St
=
BS
(USD)
Table 7: Delta hedge calculation, domestic versus foreign world.
28