Quanto Skew

Peter Jäckel∗

First version: 25th July 2009

This version: 4th June 2016

Abstract sophisticated stochastic processes for the underlying

asset.
We assess the effect of an implied volatility skew for
an FX rate on quanto forwards and quanto options 2 Common quanto adjustments
of an asset that itself is subject to an implied volatil-
ity skew using a simplistic double displaced diffusion The most commonly used quanto correction used in
model. practice is without doubt to modify the expected
value of the asset when quantoed, and otherwise reuse
whatever stochastic model may be in favour for the
underlying asset. In the case of correlated geometric
1 Introduction Brownian motion for the underlying observable and
the FX rate, it is well known that one can, by an
Quanto contracts are financial derivatives whose pay- argument of change of measure, justify the use of a
out currency differs from the natural denomination of simple adjustment for all quanto derivatives. This is
their underlying financial observable. Their purpose because simply replacing the forward F of the under-
is to provide exposure to the performance of the ob- lying by
servable without exposure to a currency conversion F 0 = F eĉ , using ĉ = σ̂S ρσ̂Q T , (1)
risk. Simple examples include forward contracts or
options on US or JPY government bonds, commod- with σ̂S being the domestic at-the-money implied
ity futures, or equity indices written such that the volatility of the asset, ρ being the process correla-
numerical return on the underlying is paid out dir- tion (i.e., correlation of increments, usually estim-
ectly in a different currency, without the application ated from time series) between asset and FX rate (in
of any FX conversion factor. This is equivalent to terms of value of one investor currency unit expressed
the concept of entering into the equivalent domestic in the asset’s domestic currency), and σ̂Q being the
derivative, e.g., a forward contract or option on a US at-the-money implied volatility of the FX rate, takes
asset written in USD, with a guarantee that whatever care of the effect of quantoing for all derivatives ex-
the payout is, it will be converted at the time of pay- piring at T on this asset. Possibly primarily for reas-
ment at an FX rate that is agreed upon at incep- ons of convenience, the quanto adjustment (1) has,
tion of the deal. Quanto contracts of vanilla nature however, also been deployed as a simple forward cor-
are traded over-the-counter in significant size. They rection irrespective of what implied volatility skew
either are requested directly by end-investors, or are may be observable for the underlying asset, and for
used as hedges for components of multi-asset invest- the FX rate.
ment strategies and other non-vanilla derivatives. When skew parametrisations are used for domestic
options on the asset in the shape of some functional
Despite their relative importance, surprisingly form as in
little research has been focussed on quanto derivat-
ive pricing. Whilst practitioners have gone to ex- σ̂(K) = f (F, K, T, λ1 , λ2 , . . .) (2)
treme lengths in their efforts to design, implement, where f () could be any functional form, e.g.,
and calibrate models for underlying assets in order to SABR [HKL02], or a volatility implied from a model
accomodate, or even explain, the observable implied expressed in terms of an underlying martingale ob-
volatility skew, when it comes to quanto derivatives, servable such as Heston [Hes93], CEV [CR76], dis-
even vanilla contracts tend to be valued with simple placed diffusion [Rub83], etc., then, in practice, the
adjustments on top of what may otherwise be a rather convential approach tends to be to retain all of the
∗
OTC Analytics
parameters λ1 , λ2 , . . . , exactly as they are used in the
Key words and phrases. quanto, skew, double displaced dif- domestic currency. Regarding the effective forward,
fusion model. there are two common schools of thought:-

1
 a) Determine an effective volatility coefficient σ̂ ex- (θ · (ST − K))+ directly in currency Y, with θ = +1
actly as in (2) using the forward as if the option for calls and θ = −1 for puts.
was entirely domestic. Subsequently, price the Denote Q as the value of one currency unit of Y
quanto option with Black’s formula replacing the expressed in currency X, i.e., as the quotient Y/X. To
forward F by F 0 as given in (1). This approach, the seller of a quanto option, the net present value of
the contract, expressed in units of currency X (e.g.,
in essence, transfers the domestic skew unaltered
for reasons of availability of options to hedge the ex-
to quanto options, whence we refer to it as posure to S), is
the Domestic-Forward-ATM-Quanto method, or
DFAQ for short. The term “ATM-Quanto” Ṽ X (t) = P (t, T ) · EX [(θ(ST − K))+ · QT ] (3)
refers to the fact that the (conventionally ap- with P (t, T ) representing today’s value of a zero
proximated) quanto forward F 0 used in the final coupon bond paying at T , and the expectation be-
pricing stage is approximated as the domestic ing taken in an X-denominated T -forward measure.
forward that is adjusted using at-the-money im- Naturally, we can translate an X-denominated net
plied volatilities. present value into a Y-denominated net present value
by spot FX conversion:
b) Determine the effective volatility coefficient dir-
Ṽ X (t)
 
Y X QT
ectly with the quanto-adjusted effective forward Ṽ (t) = = P (t, T ) · E (θ(ST − K))+ · .
Qt Qt
F 0 as given in (1). Subsequently, price the
(4)
quanto option with Black’s formula using the
same adjusted forward F 0 . In this article, we Equations (3) and (4) are model-independent.
refer to this approach as the Quanto-Forward-
ATM-Quanto, or QFAQ for short.
4 Displaced Diffusion
In this article, we attempt to shed some light on
The displaced diffusion model [Rub83] is a convenient
the question how well these simplistic quanto adjust-
device to generate a skewed implied volatility pro-
ments perform when we go beyond the Black-Scholes
file when explicit control over the curvature, i.e., the
setting, when even one of the simplest of all self-
actual smile, is of secondary importance. It allows
consistent skew-generating models is used. In our
for vanilla option valuation formulae in terms of the
investigation, we use an extremely light-weight model
Black-Scholes pricing formula with adjusted input
for the skew in the asset and the FX rate in aid of
parameters which makes it exceptionally easy and ef-
attaining closed form solutions for vanilla quanto op-
ficient in practical applications. Its main drawbacks
tions which we use for comparison. Whilst a very
are that the curvature of the implied volatility profile
simple model is clearly not enough to permit a gen-
it generates is implicitly determined by its skew, and
eral statement about the quanto effect for all models,
that the domain of the distribution it generates for
we believe it suffices to give a first order indication
the underlying financial observable does not begin at
as to the quality of conventional quanto adjustments.
zero, as for the lognormal distribution, but at an off-
This entails a closer look at the exact quanto for-
set, which is negative when the skew is negative. In
ward correction, which turns out to be model depend-
other words, when calibrated to a typical equity or
ent. Further, we attempt to give an indication of the
interest rate skew (at the money), it permits for the
quanto effect on the model skew parameters. Whilst
underlying observable to attain negative values. This
the analysis presented here is, admittedly, predicated
is clearly a deficiency in a variety of circumstances.
on very simple modelling ideas, it highlights that, in
Nevertheless, it has become a favourite of many re-
general, it is necessary to adjust the full implied volat-
searchers and practitioners whenever an assessment
ility skew when an asset is quantoed into another cur-
of the impact of a mere skew of implied volatilities
rency.
on a specific issue of financial engineering is required.
In the simple case of constant parameters, the dis-
placed diffusion model can be summarised as the
3 Exact quanto valuation unique martingale process
 1 2 2 .
Assume a financial observable S is denominated in S(t) = S(0) · e− 2 σS βS t+σS βS W (t) − (1 − βS ) βS (5)
currency X. Assume that an investor, whose nat-
ural home currency is Y, wishes to participate in for a financial observable S(t) with W (t) being a
the asset’s performance between today and some fu- standard Wiener process. European vanilla option
prices can be calculated by a simple transformation
ture time horizon T , but wishes not to be exposed of variables and are given by
to any currency risk. This investor may be inter-
(1−β)
ested in what is known as a quanto option that pays V±1 (S, K, σ, β, T ) = B±1 ( β1 S, K + β S, σβ, T ) (6)

2
  
with the Black-Scholes formula QT
= EX (±(ST − K))+ · (17)
h Qt
Bθ (S, K, σ̂, T ) = θ · S · Φ θ · ( xς̂ + 2ς̂ ) −


(7) " 
2 β2 T
σS

X S0 − S +σS βS WS (T ) 1−βS

i =E ± βS e
2 − βS S0 −K
K · Φ θ · ( xς̂ − 2ς̂ ) +
(18)
x = ln(S/K) (8) 2 β2 T
√
 σQ
 
Q
− +σQ βQ WQ (T )
ςˆ = σ̂ · T (9) · e 2 − (1 − βQ ) βQ
 
where we have omitted all discounting. At the money, = 1 1 c̃ (1−βS )
βQ B±1 βS S0 e , K + βS S0 , σS βS , T
the pricing formula (6) becomes  
(1−βQ ) 1 (1−βS )
  √   − βQ B±1 βS S0 , K + βS S0 , σS βS , T (19)
σβ T
V+1 (S, K, σ, β, T )|K=S = S · 2Φ 2 − 1 β1 (10)
with
for call options. We can compare this with the Black- c̃ = ρσS βS σQ βQ T (20)
Scholes call option price at the money given by
  √   wherein ρ represents the correlation of WS (T ) and
B+1 (S, K, σ̂(K), T )|K=S = S · 2Φ σ̂atm
2
T
− 1 (11) W Q (T ). The par strike for a quanto forward contract
can equally be computed:
with σ̂atm = σ̂(S). This allows for an explicit solution
F̃ = S0 · 1 + ec̃ − 1 /(βS βQ ) .


(21)
for the at-the-money Black implied volatility
  √   Armed with this information, we can attempt to find
(1−β)
σ̂atm = √2
T
Φ−1 1
βΦ
σβ T
2 − 2β . (12) matched quanto displaced diffusion parameters whose
purpose it is to enable the (approximate) valuation
For calibration purposes, the inverse solution is also of Y-denominated options on the underlying asset
of practical use: without the need for a combined asset-FX model.
  √   For standard geometric Brownian motion, we do of
σ = β √2 T Φ−1 βΦ σ̂atm2 T + (1−β) 2 . (13) course know how to do this change of measure exactly
— the volatility information of S in the Y measure
Of further interest is that we can derive from this and is the same as in the X measure, but the par forward
d d
changes to the quanto forward. For other models,
dK B+1 (S, K, σ̂(K), T ) = dK V+1 (S, K, σ, β, T ) (14) the change of measure results in more complicated
deformations of the quantoed asset distribution. In
the at-the-money Black implied volatility skew solely other words, not just the forward changes, but the en-
in terms of β and the at-the-money Black implied tire implied volatility profile changes, too. Here, we
volatility itself: try to approximate the quanto skew with the same
d
 √
(β−1) 2π
h  √  i σ̂2 T form of parametrisation as the domestic skew: the
dK σ̂(K) K=S =

2 S T
√ 2Φ σ̂ 2 T − 1 e 8 (15) quanto forward F̃ which is already given in equa-
  tion (21), a quanto displaced diffusion volatility σ̃,
σ̂ 2 T σ̂ 4 T 2
= (β−1)
2
σ̂
S 1 + 12 + 240 + · · · , (16) and a quanto skew parameter β̃. The hope is that,
given the right choice of the quanto skew parameters
where σ̂ = σ̂atm on the right hand side. Equa- σ̃ and β̃, we can use the vanilla displaced diffusion
tions (13) and (15) enable us to calibrate a displaced formula (6) to approximate quanto options, i.e.,
diffusion model to a given at-the-money volatility and
Y
skew with great ease. V±1(F̃ , K, σ̃, β̃, T ) ≈ Ṽ±1(S0 , K, σS , βS , σQ , βQ , ρ, T ) .
(22)

5 Displaced Diffusion Quanto

For this purpose, we match the call option price and
the skew at the quanto forward. This will ensure that
Skew at least for options struck near the quanto forward,
we will have good agreement. We shall see later how
We assume a displaced diffusion model for both the far the approximation can be used. For this, in addi-
underlying asset and the foreign exchange rate pro- tion to formula (19), we need
cess in the domestic martingale measure of the asset
(currency X) as discussed in section 3. We note that d Y
dK Ṽ+1 (S0 , K, σS , βS , σQ , βQ , ρ, T )
this, in general, prohibits us from explicitly changing    
1−β
measure to the investor’s domestic measure (currency = βQQ Φ ξς − 2ς − β1Q Φ ξ+c̃ ς
ς − 2 (23)
Y) since the reciprocal foreign exchange rate 1/Q is
not a measurable process as Q can attain zero and with
even become negative when β < 1. In this setting,  
the undiscounted quanto option price (4) is given by ξ = − ln 1 − βS + βS SK0 (24)
√
Y
Ṽ±1(S0 , K, σS , βS , σQ , βQ , ρ, T ) ς = σS βS T . (25)

3
Using the abbreviations DFAQ — for this, we use directly the simple displaced
 diffusion formula (6). Note that this is by construc-
Y tion identical to the domestic skew.
ṽ := Ṽ+1 (S0 , K, σS , βS , σQ , βQ , ρ, T ) (26)

K=F̃
QFAQ — this is the approach of replacing the for-

d Y
κ̃ := dK Ṽ+1 (S0 , K, σS , βS , σQ , βQ , ρ, T ) (27)

K=F̃ ward and reusing the domestic parameters, i.e.,
V (F 0 , K, σS , βS , T ), as explained in section 2.
we can now express our objective as solving
exact — this is formula (19).
F̃ h  1 √  i
approximate — this is the approximation
ṽ = 2Φ 2 σ̃ β̃ T − 1 (28)
β̃ V (F̃ , K, σ̃, β̃, T ) with (21) and (30).
 √ 
κ̃ = −Φ − 12 σ̃ β̃ T (29) S0
QFAQ+ — this is V (F 0 , K, σS F 0 , βS , T ).

All data are expressed in terms of implied volatilit-

for β̃ and σ̃. The solution is:- ies using the respectively exact forward as reference.
This is to mean that what is shown for any of the
β̃ = (1 + 2κ̃)F̃ /ṽ price curves is the unique solution of
 . √ (30)
σ̃ = 2 Φ−1 21 (1 + β̃ṽ/F̃ ) (β̃ T ) . vθ = Bθ (F, K, σ̂θ (vθ , F, K), T ) (31)
for σ̂ (which thereby implicitly becomes a function
of vθ , F , and K), with vθ being the respective price
6 Examples (or approximation) and F the par forward strike that
is consistent with the respective price formula. Spe-
We now give some examples for the quality of the cifically, we have:-
quanto skew approximation (30) for the double dis-
DFAQ — F = F 01
placed diffusion setup. Before we look at the quality
QFAQ — F = F0
of our approximation, we start off by highlighting
exact — F = F̃
a fact that is unfortunately far too easily overlooked approximate — F = F̃
when dealing with quanto skew parametrisations and QFAQ+ — F = F0
approximations.
Since we are using the exact forward as reference for
implied volatilities, call-put parity is preserved, and it
6.1 No FX skew is mathematically irrelevant whether we use θ = +1
This case is encompassed by our double displaced dif- or θ = −1 (though we may have chosen to always
fusion framework when βQ = 1. It is well known compute values on out-of-the money options in or-
that under these circumstances it is indeed possible der to minimize numerical truncation). Note that
to change measure to the investor currency Y since the odd choice for σS for the data in figure 1 was
the FX rate is lognormally distributed and cannot at- made such that the domestic skew is calibrated to
tain zero (or even become negative). Given that this σ̂ = 50% at K = S using equation (13). As a final
S 0
is arguably the simplest quanto framework that goes
note of form before we discuss the poignant details
beyond the skew-free pure Black-Scholes model for
the underlying asset and the FX rate, one might hope of the figure, we mention that we are not precisely
that industry-practice conventional quanto adjust- comparing like for like in the figure since we are us-
ments hold for this case exactly. We show in figure 1 ing different reference forwards for the calculation of
implied volatilities. With the exception of the DFAQ
70%
DFAQ
curve, the difference is small since, for this example,
QFAQ F 0 = 0.904837 and F̃ = 0.902336 are very close.
65%
exact
approximate The most striking effect that can be seen in figure 1
60%
QFAQ+ is that whilst the implied volatility skew of the exact
55% quanto curve appears to be shifted up, or to the right,
50%
whichever way one prefers to see it, the implied volat-
ility skew of the QFAQ quanto adjustment appears
45%
to be shifted down (or left). The discrepancy for
40% this, admittedly, or perhaps arguably2 , exaggerated
35%
1
Corrected on 2016-06-04 from S0 to F 0 , which makes this
0.6 0.8 1 1.2 1.4 1.6 consistent with the earlier statement “Note that this is by con-
Figure 1: Implied volatilities as a function of strike K for S0 = struction identical to the domestic skew ”.
1
1, T = 2, βS = 16 , σS = 48.98%, βQ = 1, σQ = 20%, and 2
A level of 50% for a two year horizon may have been con-
ρ = −50%. sidered exaggerated for equities in the pre-2008 era, but since
then much higher levels of implied volatility have been ob-
the following implied volatility curves in comparison:- served, if only temporarily.

4
example is about 5% in terms of implied volatility, with the actual price vθDFAQ computed as explained
which is clearly not negligible. The explanation for earlier in this section. Note that the apparent kink
this significant error in the conventional QFAQ ap- in the relative option time value difference curves is
proach lies with the fact that we are dealing with a no error: it indicates the location of the exact quanto
pricing framework that belongs to the family of local forward F̃ . We can see in this figure that even in this
volatility models. When we use the QFAQ quanto very simple case of a pure local volatility smile that
adjustment, we tend to focus on forward prices, and is consistent with an analytically tractable model,
too easily forget that we are effectively dealing with alas, both of the conventional quanto adjustment ap-
an approximation for a (possibly stochastic) drift in- proaches discussed here give prices that, near the ef-
troduced by the quanto term. Simply absorbing the fective money, are about 5% different from the exact
drift in an (approximately) adjusted forward, which price. We can also see that the QFAQ+ approach ap-
is then used in a formula of local volatility type has pears to eliminate most of the error incurred due to
the side effect that we inadvertently change the abso- the mis-estimation of initial absolute volatility that
lute level of volatility for the process of the underly- is inherent in the QFAQ approach. We hasten to add
ing. To lowest order, one can see this if we compare that this mis-estimation is particularly strong in the
the initial absolute volatility level of the actual as- case of a setting that is consistent with a local volatil-
set process in measure Y, which is σS · S0 , whereas ity approach. When the underlying model ideas that
the QFAQ adjustment implies that the asset process give rise to the domestic smile are closer to a float-
has initial absolute volatility close to σS F 0 . As evid- ing smile (also known as sticky delta) approach, the
ence for this effect, we included in figure 1 the curve QFAQ methodology may be more natural than the
marked as QFAQ+ where the used process volatility DFAQ approach, though, this remains to be properly
has been adjusted by the factor SF00 to compensate for investigated. Our final observation regarding figure 2
the otherwise incurred loss of volatility. As we can is that the approximate method shows zero relative
see in figure 1, this first order adjustment does indeed difference to the exact price, which is of course con-
get volatility levels almost right. Unfortunately, this sistent with this method being identical to the exact
simplistic compensation only works for moderate ma- pricing method for βQ = 1 as was mentioned earlier.
turities and choices of |βS |  1, i.e., when the model Since we consider the alternative representation of
is nearly Bachelier. As a side note, we mention that the data in terms of relative option time value errors
it can easily be shown that in this case of βQ = 1 a useful complement to the depiction of the resulting
the approximate pricing curve is identical to the ex- implied volatilities, we will from here on accompany
act solution, which is of course a desirable feature to all diagrams by this second view without further dis-
have for the approximation, and this is reflected in cussion in the text.
the data. The effect we wanted to highlight in this section
Since, as we mentioned, we are not exactly compar- is thus: when dealing with quanto skew transforma-
ing like for like in figure 1, we also show a comparison tions, there is an intrinsic risk that by focussing on
in terms of actual option price in figure 2. The data the quanto adjustment, which is often done in terms
5%
of an adjusted forward, we forget that the implied
volatility parametrisation we use, which may be in-
0% tended to be consistent with a model of local volat-
ility style, incurs absolute volatility changes if we
-5%
simply adjust the forward. The displaced diffusion
-10%
model/parametrisation used in this article is not the
only one that will have this problem. It is in fact com-
-15% mon across the whole local volatility family: CEV,
DFAQ
QFAQ
SABR, and the fully fledged non-parametric local
-20%
approximate volatility approach [Dup94]. Even when there is no
QFAQ+
-25% FX skew, vanilla quanto option pricing deserves at-
0.6 0.8 1 1.2 1.4 1.6 tention being paid to the choice of underlying model.
Figure 2: Relative option time value differences for figure 1.

in the figure are relative differences of the time value 6.2 Co-inclining skew
of the option, as priced with the respective meth- In this case, the implied volatility skew and the FX
odology, when compared with the exact price. For skew lean in the same direction. We show in figure 3
instance, the curve marked “DFAQ” corresponds to the implied volatility curves for βQ = 161
, with oth-
vθDFAQ −(θ·(F̃ −K))+ erwise the same parameters as in figure 1. As we
vθexact −(θ·(F̃ −K))+
−1 can see, the overall picture remains the same. In fig-

5
70% ure 5, we show the extreme example where we have
DFAQ
65% QFAQ increased maturity to 20 years, though we reduced
exact domestic at-the-money volatilities to 25%. We can
60% approximate
see that the discrepancies widen significantly, and
55% that our approximation (30) starts to break down
for very high strikes & 2.6, though, but agrees very
50%
well with the exact solution otherwise. It should be
45% mentioned, however, that this breakdown is largely
40%
caused by the fact that the underlying model, due
to its allowing for negative FX rates when βQ < 1,
35% in this extreme scenario, attains negative call option
0.6 0.8 1 1.2 1.4 1.6
Figure 3: Implied volatilities for S0 = 1, T = 2, βS = 16 , σ̂S = prices for & 2.77. Thus, arguably, the failure of the
1
1
50% → σS = 48.98%, βQ = 16 , σ̂Q = 20% → σQ = 19.95%, approximation, which is designed to disallow negative
0
and ρ = −50%. F = 0.9048 and F̃ = 0.9024. option prices, is of no concern greater than the funda-
mental shortcoming of the chosen simplistic model to
5% allow for negative option prices in extreme scenarios.
0%
DFAQ
55% QFAQ
-5% exact
approximate
-10%
50%
-15%

DFAQ 45%
-20%
QFAQ
approximate
-25%
0.6 0.8 1 1.2 1.4 1.6 40%
Figure 4: Relative option time value differences for figure 3.
0.6 0.8 1 1.2 1.4 1.6
Figure 7: Implied volatilities for S0 = 1, T = 2, βS = 32 , σ̂S =
80%
DFAQ 50% → σS = 51.42%, βQ = 23 , σ̂Q = 20% → σQ = 20.08%,
70% QFAQ and ρ = −50%. F 0 = 0.9048 and F̃ = 0.9079.
exact
60% approximate

14%
50% DFAQ
12% QFAQ
40% approximate
10%
30%
8%
20% 6%

10% 4%
0.3 0.4 0.5 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.3 2.6 3
1
Figure 5: Implied volatilities for S0 = 1, T = 20, βS = 16 , 2%
1
σ̂S = 25% → σS = 23.76%, βQ = 16 , σ̂Q = 20% → σQ =
0%
19.95%, and ρ = −50%. F 0 = 0.6065 and F̃ = 0.5405.
-2%
0.6 0.8 1 1.2 1.4 1.6
60% Figure 8: Relative option time value differences for figure 7.
DFAQ
40% QFAQ
approximate In figures 7 and 9, we show the example of positive
20% skew for the underlying asset and the FX rate with
0%
βS = βQ = 32 , T = 2, and ρ = −50% and ρ = 50%,
respectively. This is followed by a long dated example
-20% with positive skew in figure 11.
-40%
6.3 Contra-inclining skew
-60%

-80%
In this case, the implied volatility skew and the FX
0.3 0.4 0.5 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.3 2.6 3 skew lean in opposite directions.
Figure 6: Relative option time value differences for figure 5.

6
 60%
58% DFAQ
DFAQ
56% QFAQ
QFAQ
exact
54% exact 55% approximate
approximate
52%
50%
50%
48%
46%
44% 45%
42%
40%
40%
38% 0.6 0.8 1 1.2 1.4 1.6
0.6 0.8 1 1.2 1.4 1.6
3 Figure 13: Implied volatilities for S0 = 1, T = 2, βS = 21 , σ̂S =
Figure 9: Implied volatilities for S0 = 1, T = 2, βS = σ̂S = ,
2
50% → σS = 49.22%, βQ = 32 , σ̂Q = 20% → σQ = 20.08%,
50% → σS = 51.42%, βQ = 23 , σ̂Q = 20% → σQ = 20.08%,
and ρ = −50%. F 0 = 0.9048 and F̃ = 0.9047.
and ρ = 50%. F 0 = 1.1052 and F̃ = 1.1163.
2%
5% DFAQ
0% QFAQ
0% approximate
-5% -2%
-10% -4%
-15%
-6%
-20%
-25% -8%
-30%
-10%
-35%
-40% DFAQ -12%
QFAQ
-45% -14%
approximate
-50% 0.6 0.8 1 1.2 1.4 1.6
0.6 0.8 1 1.2 1.4 1.6 Figure 14: Relative option time value differences for figure 13.
Figure 10: Relative option time value differences for figure 9.
60%
DFAQ
35% QFAQ
DFAQ exact
QFAQ 55% approximate
30% exact
approximate

25% 50%

20% 45%

15%
40%
0.6 0.8 1 1.2 1.4 1.6
10% Figure 15: Implied volatilities for S0 = 1, T = 2, βS = 21 , σ̂S =
0.3 0.4 0.5 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.3 2.6 3
50% → σS = 49.22%, βQ = 32 , σ̂Q = 20% → σQ = 20.08%,
Figure 11: Implied volatilities for S0 = 1, T = 20, βS = 32 ,
and ρ = 50%. F 0 = 1.1052 and F̃ = 1.1026.
σ̂S = 25% → σS = 27.05%, βQ = 32 , σ̂Q = 20% → σQ =
20.96%, and ρ = 50%. F 0 = 1.6487 and F̃ = 2.1471. 14%
DFAQ
12% QFAQ
100% approximate
10%
50%
8%
0%
6%
-50%
4%
-100%
2%
-150%
0%
-200%
DFAQ -2%
-250% QFAQ 0.6 0.8 1 1.2 1.4 1.6
approximate Figure 16: Relative option time value differences for figure 15.
-300%
0.3 0.4 0.5 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.3 2.6 3
Figure 12: Relative option time value differences for figure 11.

7
60%
DFAQ
50%
DFAQ
QFAQ
45% QFAQ
exact
55% exact
approximate
40% approximate

35%
50%
30%

45% 25%

20%
40%
0.6 0.8 1 1.2 1.4 1.6 15%
0.3 0.4 0.5 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.3 2.6 3
Figure 17: Implied volatilities for S0 = 1, T = 2, βS = 23 , σ̂S =
Figure 21: Implied volatilities for S0 = 1, T = 20, βS = 12 ,
50% → σS = 51.42%, βQ = 12 , σ̂Q = 20% → σQ = 19.95%,
σ̂S = 25% → σS = 24.04%, βQ = 32 , σ̂Q = 20% → σQ =
and ρ = −50%. F 0 = 0.9048 and F̃ = 0.9047.
20.96%, and ρ = −50%. F 0 = 0.6065 and F̃ = 0.5804.
25%
DFAQ 10%
QFAQ
20% approximate 0%

15% -10%
-20%
10%
-30%

5% -40%
-50%
0%
-60% DFAQ
-5% -70% QFAQ
0.6 0.8 1 1.2 1.4 1.6 approximate
Figure 18: Relative option time value differences for figure 17. -80%
0.3 0.4 0.5 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.3 2.6 3
Figure 22: Relative option time value differences for figure 21.
60%
DFAQ
QFAQ 50%
exact DFAQ
55% approximate QFAQ
45%
exact
40% approximate
50%
35%

45% 30%

25%

40% 20%
0.6 0.8 1 1.2 1.4 1.6
3
Figure 19: Implied volatilities for S0 = 1, T = 2, βS = σ̂S = 2
, 15%
50% → σS = 51.42%, βQ = 12 , σ̂Q = 20% → σQ = 19.95%, 0.3 0.4 0.5 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.3 2.6 3
and ρ = 50%. F 0 = 1.1052 and F̃ = 1.1066. Figure 23: Implied volatilities for S0 = 1, T = 20, βS = 12 ,
σ̂S = 25% → σS = 24.04%, βQ = 32 , σ̂Q = 20% → σQ =
5% 20.96%, and ρ = 50%. F 0 = 1.6487 and F̃ = 1.6123.

0%
DFAQ
-5% QFAQ
150% approximate

-10%

-15% 100%

DFAQ
-20%
QFAQ 50%
approximate
-25%
0.6 0.8 1 1.2 1.4 1.6
Figure 20: Relative option time value differences for figure 19. 0%
0.3 0.4 0.5 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.3 2.6 3
Figure 24: Relative option time value differences for figure 23.

8
35% 100%
DFAQ DFAQ
QFAQ QFAQ
30% exact approximate
approximate 50%
25%

20% 0%

15%
-50%
10%

5% -100%
0.3 0.4 0.5 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.3 2.6 3 0.4 0.5 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.3 2.6 3
3
Figure 25: Implied volatilities for S0 = 1, T = 20, βS = 2
, Figure 28: Relative option time value differences for figure 27.
σ̂S = 25% → σS = 27.05%, βQ = 12 , σ̂Q = 20% → σQ =
19.51%, and ρ = −50%. F 0 = 0.6065 and F̃ = 0.5642.
in contrast to the most commonly used practices for
the valuation of vanilla quanto options. For the spe-
300%
DFAQ cific example of a double displaced diffusion model,
QFAQ we have given the explicit pricing formula, and de-
250% approximate
rived an analytical approximation that permits to
200% represent the quanto option implied volatility skew
in the same parametric setting as the domestic skew,
150%
within this model. We have given numerical examples
100% for a variety of parameter combinations that appear
to be in support of the practical viability of the ap-
50%
proximation. Notwithstanding the fact that the ana-
0%
lysis employed a double displaced diffusion model, we
0.4 0.5 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.3 2.6 3 emphasize that the main result of the investigation
Figure 26: Relative option time value differences for figure 25. is that quanto options, for anything but short-dated
contracts, warrant explicit modelling for accurate pri-
35% cing and consistent risk managament with respect to
DFAQ
QFAQ
the underlying vanilla markets.
30% exact
approximate
25% References
[CR76] J. C. Cox and S. A. Ross. The valuation of options for al-
20% ternative stochastic processes. Journal of Financial Eco-
nomics, 3:145–166, March 1976.
15%
[Dup94] B. Dupire. Pricing with a Smile. Risk, 7(1):18–20, 1994.
10%
[Hes93] S. L. Heston. A closed-form solution for options with
stochastic volatility with applications to bond and cur-
5% rency options. The Review of Financial Studies, 6:327–
0.3 0.4 0.5 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.3 2.6 3 343, 1993.
Figure 27: Implied volatilities for S0 = 1, T = 20, βS = 32 ,
σ̂S = 25% → σS = 27.05%, βQ = 12 , σ̂Q = 20% → σQ = [HKL02] P. Hagan, D. Kumar, and A. S. Lesniewski. Managing
19.51%, and ρ = 50%. F 0 = 1.6487 and F̃ = 1.6474. Smile Risk. Wilmott Magazine, pages 84–108, September
2002.

Figures 13 to 27, which are for various contra- [Rub83] M. Rubinstein. Displaced diffusion option pricing. Journal
of Finance, 38:213–217, March 1983.
inclining combinations of βS < 1 and βQ > 1, and
βS > 1 and βQ < 1, are explained in their respective
captions.

7 Conclusion
We have shown that the comparatively simple finan-
cial derivative contract known as a “quanto” option
requires model-dependent considerations when both
the underlying asset and the converting FX rate’s op-
tion markets exhibit implied volatility skews. This is