FX correlation swap: problemsolver or troublemaker?

Alvise De Col and Patrick Kuppinger∗

December 19, 2017

Abstract
We show that FX correlation swap prices show a rich and non-trivial dependency on often overlooked higher
order parameters such as the correlations between the FX variances. As a consequence, missing reliable information
to calibrate these quantities may result in a non-negligible uncertainty in the corresponding quotes. We first review
the industry-typical choice of multi-dimensional FX models (Black-Scholes and local correlation) and then show that
stochastic local correlation models are capable to capture a wide spectrum of the uncertainty range in FX correlation
swap quotes.

1 Introduction
Multi-asset products have become fairly standard for derivative investors seeking a diversified exposure or premium
rebates by warehousing correlation risk. Differently to other asset classes, in FX a significant amount of information
about the correlations can be obtained directly from the market through triangulation of the FX volatility surfaces
[1–7]. In this paper, we put focus on the most natural correlation product, the FX correlation swap. This structure
has become largely popular among investment banks looking to hedge their existing correlation exposure within
their multi-dimensional FX books. Moreover, buy-side firms—mainly hedge-funds—show appetite in such products
enabling them to express a direct view on correlation moves.
To describe an FX correlation swap, we consider three currencies CCY 0, CCY 1, and CCY 2, where CCY 0
denotes the numeraire currency. This yields two main currency pairs S1 = CCY 0/CCY 1 and S2 = CCY 0/CCY 2,
respectively, and a cross currency pair S3 = CCY 1/CCY 2 = S2 /S1 (where the notation CCY i/CCY j means the
amount of currency i for one unit of currency j). The payoff of a correlation swap with unit notional is defined as
cov1,2
P (S1 , S2 ) = −K (1)
Σ1 Σ2
where K denotes the strike. The covariance and realized volatilities of S1 and S2 are defined as
n    
An X S1 (ti ) S2 (ti )
cov1,2 = ln ln
n−1 S1 (ti−1 ) S2 (ti−1 )
i=2
√
and Σj = covj,j , respectively. Above, An stands for the annualization factor, and n is the number of fixings at
times {ti } with ti = ti−1 + ∆t. Sometimes, the correlation swap is defined including a mean adjustment, which does
not affect the results below.
A solid pricing and risk management framework capable of describing the joint dynamics of the underlying ex-
change rates is needed to consistently price and hedge such a multi-dimensional FX derivative. One key requirement—
highlighted in [6,7]—is to use a model that respects the inversion and triangulation symmetries inherent to exchange
rates. At the same time, any multi-dimensional FX model should consistently price vanilla options written on the
individual underlyings. In what follows, we shall assume a multi-dimensional FX model describing the joint dynamics
of the two main currency pairs S1 and S2 such that above mentioned symmetry properties are satisfied, and vanillas
∗ The authors are with UBS Business Solutions AG, a subsidiary of UBS Group AG. The article contains personal views expressed by the

authors and may not reflect the views of UBS Group AG or any of its subsidiaries.

Electronic copy available at: https://ssrn.com/abstract=3228827

on the mains and the cross are priced consistently. Several choices for such a model are described in [7]. If we
consider the one-dimensional projections of the chosen model, then Sj can be described by an SDE of the form

σj2 (t)
 
d ln Sj (t) = mj (t) − dt + σj (t)dWj
2

where mj denotes the risk neutral drift of Sj in the CCY0 measure, and σj denotes the instantaneous volatility of Sj
(note that m3 contains a quanto drift adjustment). Importantly, the instantaneous volatilities shall not be further
specified; in particular, they can be stochastic processes themselves. In this paper, we consider the continuous limit
∆t → 0 (discretization effects are therefore neglected), in which case the correlation payout (1) becomes
RT
ρS ,S (t)σ1 (t)σ2 (t)dt
Pcont = q0 R 1 2 − K. (2)
T 2 RT
0
σ1 (t)dt 0 σ22 (t)dt

In (2), ρS1 ,S2 (t) denotes the instantaneous correlation between the Brownian drivers of S1 and S2 . Note that the
correlation in general might depend on S1 , S2 , and also other quantities present in the model (e.g., stochastic
volatilities). Looking at S2 /S1 , the instantaneous volatility σ3 of the cross is linked to the correlation via the
triangular relationship:

σ32 (t) = σ12 (t) + σ22 (t) − 2ρS1 ,S2 (t)σ1 (t)σ2 (t).

Hence, we can rephrase (2) in terms of the instantaneous volatilities of the two mains and the cross currency pair:
RT 2
(σ1 (t) + σ22 (t) − σ32 (t))dt
Pcont = 0 qR − K.
T 2 RT 2
2 0 σ1 (t)dt 0 σ2 (t)dt

Taking the expectation in the CCY0 risk-neutral measure, we obtain the fair strike of the correlation swap:
" #
1 V1 (T ) + V2 (T ) − V3 (T )
Kfair = E0 p (3)
2 V1 (T )V2 (T )
RT
where Vj (T ) = 1/T 0 σj2 (t)dt denotes the annualized realized variance of the jth underlying within the generic
model.
In the remainder of the paper, we will further investigate the fair correlation strike expression (3). In particular,
we shall discuss how this quantity depends on the volatility dynamics and their correlations.

2 Decomposition of the Fair Correlation Strike

Let us next gain a qualitative understanding of the behavior and dependence of the fair strike expression (3) on
the observable market and also the chosen model parameters. Assume that the realized variances each follow some
arbitrary distribution with mean E0 [Vj ] = µj and variance Var0 [Vj ] = α2j , which means that the volatility of realized
variance of the jth FX rate within the model is given by αj . Furthermore, we shall assume that all higher moments of
the realized variance are small. Here, it is important to note that we take expectation under the CCY0 risk-neutral
measure. While for the two main FX rates this means that µ1 and µ2 correspond to the fair variance swap strike
on S1 and S2 , respectively, for the cross it means that µ3 corresponds to the fair quanto variance swap strike, i.e.,
the fair strike of a variance swap paying in a different currency than its natural domestic currency. In this case, the
natural payoff currency for the variance swap on the cross would be CCY1. However, since we take expectation
under the CCY0 risk-neutral measure, it corresponds to a variance swap on the cross quanto’ed in CCY0. This
subtlety will be discussed in more detail towards the end of this section.
We begin the analysis by splitting the fair correlation strike into three components:
"s # "s # " #!
1 V1 (T ) V2 (T ) V3 (T )
Kfair = E0 + E0 − E0 p . (4)
2 V2 (T ) V1 (T ) V1 (T )V2 (T )

Electronic copy available at: https://ssrn.com/abstract=3228827

Assuming that volatilities and moments of order three or larger of the realized variance are small we can make the
following Taylor series expansion of the individual terms in (4):
"s # " ! !#
V1 (T ) √ V 1 − µ1 (V1 − µ1 )2 1 V2 − µ2 3(V2 − µ2 )2
E0 ≈ E0 µ1 + √ − p √ − p 3 + p
V2 (T ) 2 µ1 8 µ31 µ2 2 µ2 8 µ52
α2 3α22
r  
µ1 ρV1 ,V2 α1 α2
= 1 − 12 + − + O(α3 ) (5)
µ2 8µ1 8µ22 4µ1 µ2

where we have used E[(V1 − µ1 )(V2 − µ2 )] = ρV1 ,V2 α1 α2 , ρV1 ,V2 being the correlation between the realized variances
V1 and V2 . Similarly, we can decompose the other two terms in (4) and eventually obtain the following expression for
the fair correlation strike as a function of the fair variance swap strikes µ1 and µ2 of the two main FX rates, the fair
quanto variance swap strike µ3 of the cross FX rate, the volatilities of the realized variances, and the correlations
between the three variance processes:

µ1 + µ2 − µ3 (µ1 − 3µ2 + 3µ3 )α21 (µ2 − 3µ1 + 3µ3 )α22 (µ1 + µ2 + µ3 )ρV1 ,V2 α1 α2
Kfair ≈ √ − √ − √ − √
2 µ1 µ2 16µ21 µ1 µ2 16µ22 µ1 µ2 8µ1 µ2 µ1 µ2
 
1 ρV1 ,V3 α1 α3 ρV ,V α2 α3
+ √ + 2 3 . (6)
4 µ1 µ2 µ1 µ2

We recognize the first term as the fair strike of the correlation swap in terms of fair (quanto) variance swap strikes,
in absence of stochastic volatility. All the other terms describe the convexity adjustment due to the stochastic
variance processes and their correlations. Furthermore, we observe a short exposure of the fair strike on the main-
main variance correlation ρV1 ,V2 and a long exposure on the main-cross variance correlations ρV1 ,V3 and ρV2 ,V3 . This
long/short dependency on the variance correlations produces a rather rich pricing behavior as will be shown further
below.
By means of (6) we analyse the fair strike in various industry standard models that differ in the volatilities of
realized variance and the correlations among the variance processes.

2.1 Correlation Swap Pricing in Various Models

We consider a set of typical multi-dimensional FX models: (multi-dimensional) Black-Scholes (BS) [1], local corre-
lation (LC) [2, 7, 8], and stochastic (local) correlation (SLC) [2, 6–8].
In order to focus on the key role played by the correlations between the variances we shall assume throughout
this section that µj ≈ µ, ∀j, which—for example—will be the case if the three FX implied volatility surfaces are
identical (ignoring the quanto effect on the cross that we will address further below in Section 2.2). It is otherwise
straightforward to work with the more general expression in (6) if needed. The fair strike in (6) thus simplifies to
1 1
α21 + α22 + 6ρV1 ,V2 α1 α2 − 4ρV1 ,V3 α1 α3 − 4ρV2 ,V3 α2 α3 .


Kfair ≈ − (7)
2 16µ2
Expression (7) lends itself nicely to understanding the key drivers behind correlation swap pricing. While some
simplifying assumptions were made in order to arrive at (7), numerical test results presented further below confirm
its prediction power.

2.1.1 Black-Scholes
For BS we have αj = 0; thus, (7) yields
1
Kfair (BS) = . (8)
2
Clearly, the BS model is overly simplistic to adequately price a product such as the FX correlation swap.

2.1.2 Local Correlation

In an LC model we have nonzero values for the volatilities and the correlations of realized variances due to the
spot-dependency of the local volatilities. However, the volatilities of realized variances are usually smaller in LC

Electronic copy available at: https://ssrn.com/abstract=3228827

Figure 1: Left plot: Correlation between two realized variances (common driver case) and variance of realized variance
as a function of the mixing weight (note that mixing weight equal to zero corresponds to a local volatility model,
mixing weight equal to one corresponds to a volatility of variance fully calibrated to the vanilla smile; by varying the
mixing weight, one can achieve all intermediate values of the volatility of variance). Right plot: Implied volatility fit to
the market for one main and the cross currency pair (based on an artificially chosen FX market with three identical
implied volatility surfaces). Both plots are shown for an expiry of one year; a Heston process was used for the stochastic
volatility part (common driver case).

Electronic copy available at: https://ssrn.com/abstract=3228827

models compared to models including an explicit stochastic volatility component [9], see Fig. 1. For the LC model
with identical implied volatility surfaces we obtain the following expression for the fair strike of the correlation swap:
1 α2
Kfair (LC) ≈ − LV2 (1 − ρLV ). (9)
2 8µ
Hence, we expect the LC model to have a slightly lower fair strike of the correlation swap than the BS model. Note
that the volatility of realized variance αLV and the correlation ρLV in an LC model are not model inputs and instead
are enforced by the local volatility surfaces (which cannot be controlled by the user). Similarly to the BS model,
the LC model, despite of its additional complexity, still lacks degrees of freedom to capture the uncertainty in FX
correlation swap prices as indicated by (7).

2.1.3 Stochastic Local Correlation

Let us next turn our focus on stochastic local correlation (SLC) models. In the literature there exist several de-
scriptions of such models that may differ in the stochastic variance processes, the correlation structure between
those processes, and the amount of stochastic volatility in the model [2, 7, 8]. As already mentioned SLC models
usually have larger volatilities of realized variance as their LC counterpart, while still matching all implied volatility
surfaces. The amount of volatility of realized variance can be controlled by the volatility of variance or by the mixing
weight [1], if such an approach is taken in the calibration of the underlying stochastic volatility models. The mixing
weight approach allows the user to fade the model between a purely local volatility model at mixing weight zero and
a predominantly stochastic volatility model with some local volatility addon at mixing weight one. The correlation
between the realized variances can be influenced by the user’s choice of correlations between the stochastic variance
drivers. Note, however, that neither the volatilities nor the correlations among realized variances can be fully chosen
by the user as the remaining local volatility component will additionally impact those quantities. Nevertheless, we
can demonstrate that the user has a fairly direct handle on the volatility and correlation and can access—differently
from the models analyzed so far—a wider range of fair correlation swap prices, as predicted by (7).

SLC with Symmetric Setup First, we consider a full SLC model in a symmetric setup (i.e., all variance
correlations and all volatilities of variances equal, αj = α, ρVi ,Vj = ρ), thus yielding

1 α2
Kfair (symmetric SLC) ≈ − (1 − ρ). (10)
2 8µ2
This expression is formally the same as we obtained for the LC model in (9). However, as already pointed out and
visualized in Fig. 1, the volatilities of realized variances and the correlations between those variance processes in the
SLC model are independent model inputs and can be modified by the user. As a consequence, larger values for α
can be achieved. Therefore, the maximum reduction of fair strike versus the value observed in BS is expected to be
significantly larger in SLC than in LC.

SLC with Common Driver In [2, 7] it is advised to use SLC models with one common variance driver—mainly
for stability reasons. This in turn would mean that the correlations between the realized variances are close to one
for a large mixing weight. For a mixing weight of zero, LC and SLC common driver are equivalent. Increasing the
mixing weight towards one, two effects can be observed (see Fig. 1) in (10): α increases, but also ρ increases
(for increased mixing weight, we get less contributions coming from the local volatility part in the model, hence
correlations between the realized variances are closer to one). These two effects are offsetting each other in (10),
and as shown numerically in Section 3 only small deviations are observable between the fair strikes produced in LC
and in SLC with common driver. This is an artefact of the specific choice of correlation between variance drivers.
If only this version of SLC is considered, one might arrive at the wrong conclusion that correlation swaps are fairly
model independent. Hence, one sees the need to free up the correlations between the variance drivers.

SLC with Independent Drivers Another special case of SLC with symmetric setup is the one where we assume
independence between the drivers of the stochastic variance processes (αj = α, ρVi ,Vj ≈ 0). This will result in the
following expression for the fair strike:
1 α2
Kfair (SLC independent drivers) ≈ − (11)
2 8µ2

Electronic copy available at: https://ssrn.com/abstract=3228827

which is implying a reduced value for the fair strike compared to the BS model. Note that the reduction is proportional
to the volatility of the realized variance. Hence, varying the mixing weight will allow the user to control the amount
of reduction versus the BS value.

SLC with Local Cross A typical simplification in SLC models is to assume a purely local diffusion for the cross
FX rate. In such a setup the volatility of realized variance for the mains outweighs the one for the cross so that we
may ignore the terms containing α3 . In this configuration (α1 = α2 = α, α3  α) we obtain

1 α2
Kfair (SLC local cross) ≈ − (1 + 3ρV1 ,V2 ) . (12)
2 8µ2
Note that within this model choice, depending on how one sets the correlation between the stochastic variance
drivers of the two mains (and the mixing weights of the mains), the fair correlation swap strike may show an even
more pronounced reduction versus the BS value than in the previously discussed cases.

SLC with Local Main The cases we have investigated above have in common that they lead to a reduction
in fair strike versus the BS model (assuming positive realized variance correlation). It is also possible to achieve an
increase in fair strike. For example, one might choose to have a small volatility of variance for one of the mains
(e.g., by switching off the stochastic volatility part of the model on that main by reducing the mixing weight). This
setup with perfectly correlated variance processes (α2 = α3 = α, α1  α, ρV2 ,V3 ≈ 1) yields

1 3α2
Kfair (SLC local main) ≈ + . (13)
2 16µ2

2.2 The Quanto Effect on the Cross FX Variance

In order to compute the fair strike of the correlation swap we take the expectation in the CCY0 risk-neutral
measure, which is not the natural measure for the cross FX rate. As a consequence µ3 is not the standard fair
variance swap strike on the cross FX rate (that is determined fully by the corresponding implied volatility surface)
but the fair quanto variance swap strike. Assuming zero rates, the standard variance swap on the cross is given by
E1 [V3 ] = E0 [S1 (T )V3 ]/S1 (0). The quanto variance and the variance swap are related by S1 (0) E0 [V3 ] = E0 [S1 (T )V3 ]−
E0 [(S1 (T )−S1 (0))V3 ], which means that the quanto variance swap is equal to the variance swap minus the covariance
between S1 and V3 . The leading term of this covariance depends on the correlation between S1 and V3 , the volatility
of S1 , and the volatility of V3 (the argument does not change for non-zero rates). More specifically, we can write
√
cov0 (S1 , V3 ) ≈ ρS1 ,V3 µ1 S1 (0)α3 + O(α23 ). In particular, if either the correlation between S1 and V3 or any of the
√
two volatilities (i.e., spot volatility µ1 or volatility of variance α3 ) is zero, the quanto variance swap is equal to the
variance swap. Note that for SLC models in FX it is often the case that the correlation between the spot and the
variance drivers is set to zero (in order to preserve the model symmetry under inversion and triangulation [2]). In this
case, as the correlation between the drivers can be seen as a proxy for the correlation between the corresponding
realised quantities, we expect the quanto effect to be negligible.

3 Test Results
In this section we present a selection of tests linking the theoretical findings to actual correlation swap pricing results.
We consider the types of models that we discussed in Section 2.1. All tests cases were obtained with stochastic
volatility processes of Heston type like in [7] but allowing for arbitrary correlation structure among the variance
drivers. As described above, to preserve the FX symmetries, the correlation between spot and variance drivers is set
to zero. All numerical results are based on Monte Carlo valuations using 40K paths.
In consistency with (7), for the test results shown in Fig. 2 we concentrate on the key components impacting
correlation swap prices. Therefore, we consider an artificially chosen FX market with three identical implied volatility
surfaces with nonzero skew and kurtosis (see Table 1 for an example expiry of one year). The fair variance swap
strikes µj are the same for the two main FX rates. However, for the cross, µ3 denotes the fair quanto variance
swap strike and therefore contains a correction with respect to the variance swap as we have discussed above.
Nevertheless, for the test cases we considered, the quanto impact on the fair correlation swap strike was below one
correlation point and, hence, does not fundamentally alter the theoretical predictions made in the previous section.

Electronic copy available at: https://ssrn.com/abstract=3228827

Figure 2: Comparison of the one-year correlation swap fair strike for BS, LC, and various versions of SLC: SLC common
driver, SLC independent drivers (11), SLC local cross independent mains (12) with ρV1 ,V2 = 0, SLC local cross common
main driver (12) with ρV1 ,V2 = 1, and SLC local main common driver (13). As explained in the text, the mixing weight
applies only to those underlyings that are not fully local. Here, we consider an artificially chosen FX market with three
identical implied volatility surfaces (see Table 1).

Electronic copy available at: https://ssrn.com/abstract=3228827

 10DP 25DP ATM 25DC 10DC µ3 var swap on S3
Market 12.18 11.11 10.11 9.66 9.76
SLC local main common driver 12.32 11.30 10.34 9.81 9.81 0.01139 0.01155
SLC common driver 12.17 11.10 10.13 9.70 9.78 0.01142 0.01155
LC 12.16 11.09 10.12 9.67 9.75 0.01149 0.01161
SLC independent drivers 11.91 10.99 10.19 9.83 9.96 0.01145 0.01157
SLC local cross independent mains 12.08 11.08 10.17 9.80 9.94 0.01152 0.01163
SLC local cross common main driver 11.78 10.79 9.92 9.55 9.64 0.01145 0.01156

Table 1: Implied volatility fit for the cross currency pair given the market data used for Fig. 2 for a maturity of one year.
We consider the worst-case scenario (for vanilla fitting) where mixing weight is equal to one; numbers (except the var
swap fair strike) are in volatility points. The variance swap fair strike on the mains is 0.0116. From the differences with
this value, we may conclude that the impact on the fair correlation swap strike is consistently below one correlation
point for all the models considered in the table.

Clearly, for the BS model we have amongst others also that α3 = 0 and therefore no quanto correction. As a
consequence, we obtain—as predicted in the theoretical section in (8)—a fair strike of the correlation swap of 1/2
(50 correlation points).
The LC model corresponds to a scenario where all volatilities of variances are equal and small due to the lack of
stochastic volatility. The LC results in Fig. 2 correspond to what was predicted in (9). In particular, we observe that
LC falls slightly below the BS results as a consequence of the small negative contributions arising from the nonzero
volatilities of realized variances.
We also present results for a variety of different flavours of SLC models corresponding to the theoretical cases
we discussed above. In this particular test case we observe that a symmetric SLC model spans a correlation regime
of around five correlation points for a mixing weight of one with the two extreme cases common and independent
variance drivers (see (10) and (11)). Moreover, we observe that LC and SLC with common driver match very well,
as predicted above. In case the symmetry of the model is broken and part of it is assumed to be local (local cross
or local main, for example), then a large impact on the realized correlation can be seen—as predicted by (12) and
(13). The difference in fair strike can easily span a range of around 25 correlation points.
In Fig. 3 we show the dependence of the fair strike of a correlation swap valued under SLC in a symmetric setup
on the mixing weight (i.e., volatility of realized variance) and on the correlation among the variance drivers (i.e.,
correlation among the realized variances). For these results, real market data was used (i.e., the three implied volatility
surfaces were not identical), and one can observe that (10) describes the behavior well: for a given correlation, the
fair strike is a decreasing function of the mixing weight; for a given mixing weight, the fair strike is an increasing
function of the correlation. In the fully correlated setup (SLC with common driver), we see that the fair strike now
has a dependency on the mixing weight (however, still the least of all considered cases). This is due to the fact that
for real market data the terms in (6) do not fully offset each other like they do in the simplified case we considered
in Section 2.1.
As discussed in [7], if one allows for a general correlation structure among the variance drivers, more violations of
the required positive-semidefiniteness of the spot correlation matrix are expected. This, in turn, requires an increased
number of corrections, which will ultimately deteriorate the vanilla fit on the cross. For the test cases above it was
verified that the vanilla fit on the mains and the cross remained acceptable (see Table 1 and Table 2 for the cross;
the vanilla fit error for the main currency pairs was consistently below 0.1 vol points and is therefore not shown in
the tables). However, the results in this paper do not rely on the calibration quality of individual points on the vanilla
surfaces but rather on consistent pricing of variance swap fair strikes, which condense the required information
about all strikes including wing behavior into one number. Given that the calibration of the mains is unproblematic,
any potential calibration issues are hence due to misfits of the variance swap fair strike on the cross. In Table 1
and Table 2 we show that this is preserved within the required accuracy to guarantee robustness of the numerical
results. Moreover, as already mentioned previously, the impact of quantoing is equally small. In order to quantify
the potential impact of the calibration error one can look at the leading term in (6), yielding that the effect on the
√
fair correlation swap strike is given by ∆Kfair ≈ −∆µ3 /(2 µ1 µ2 ). For the results obtained, this was less than one
correlation point.

Electronic copy available at: https://ssrn.com/abstract=3228827

Figure 3: Dependence of the fair correlation swap strike in a one-year EURUSD-GBPUSD correlation swap valued
under SLC with symmetric model setup on the mixing weight and the correlation among the variance drivers (αj =
α, ρVi ,Vj = ρ). The market data snapshot is from December 2017.

10DP 25DP ATM 25DC 10DC µ3 var swap on S3

SLC common driver 0.04 0.04 0.03 0.02 -0.01 0.00729 0.00732
SLC independent drivers 0.12 0.24 0.42 0.32 0.15 0.00727 0.00731
LC 0.02 0.01 0.01 0.00 -0.04 0.00733 0.00735

Table 2: Difference in vol points of the implied volatility for the cross currency pair given the market data used for
Fig. 3 and a maturity of one year. We consider the worst-case scenario (for vanilla fitting) where mixing weight is equal
to one. From the differences between SLC and LC in the fair (quanto) variance swap strikes we may conclude that the
impact on the fair correlation swap strike is consistently below one correlation point.

Electronic copy available at: https://ssrn.com/abstract=3228827

4 Final Remarks
This study carves out the importance of the correlation structure among realized variances on the pricing of products
like the correlation swap. The formalism is agnostic with respect to the particular choice of model and is capable to
fully explain the seemingly intricate pricing behavior.
As was shown with above results, a wide range of correlation swap prices can be achieved with just one family
of SLC models by modifying the mixing weights and the correlations between variance drivers. Clearly, not all
configurations of parameters make sense other than from a theoretical point of view. SLC models require additional
calibration effort compared to LC models. In particular, not only a sound calibration of the mixing weights but also
the correlations between the stochastic variance drivers is important to avoid massive uncertainty in the pricing of
products like correlation swaps.
Typically, mixing weights are calibrated on the individual (single-dimensional) components of the model by
matching certain liquid path dependent options such as barriers or double no-touches [1]. The calibration of the
correlation structure among the stochastic variance processes is not much discussed in the literature, as there
are not as many liquidly traded multi-dimensional products in the market that yield enough information about
these parameters. One way to calibrate the correlation could be to use information embedded in firmly quoted
multi-dimensional baskets or, alternatively, historical series of implied volatilities. Missing reliable information on the
correlation between variance processes, trading needs to incorporate the uncertainty due to this usually overlooked
model parameter in correlation swap quotes.

Acknowledgement
The authors would like to thank Stephane Matar for several interesting and useful discussions on the topic of this
paper while he was at UBS AG and Manos Venardos for his insightful comments.

References
[1] I. J. Clark, “Foreign exchange option pricing: A practitioner’s guide,” John Wiley & Sons, Ltd, 2011.
[2] P. Austing, “Smile pricing explained,” Palgrave Macmillan, 2014.
[3] P. Carr and D. Madan, “Introducing the covariance swap,” Risk, Feb. 1999.
[4] A. Langnau, “A dynamic model for correlation,” Risk, Mar. 2010.
[5] A. Campbell-Smith and B. Hamdani, “The rise of multi-currency options,” Risk, Aug. 2010.
[6] A. De Col, A. Gnoatto, and M. Grasselli, “Smiles all around: FX joint calibration in a multi-Heston
model,” Journal of Banking and Finance, no. 37, pp. 3799–3818, 2013.
[7] A. De Col and P. Kuppinger, “Pricing multi-dimensional FX derivatives via stochastic local correla-
tions,” Wilmott magazine, no. 73, pp. 72–77, 2014.
[8] J. Guyon, “Local correlation families,” Risk, Jan. 2014
[9] J. Gatheral, “The volatility surface, a practioner’s guide,” Wiley Finance, 2006.

10

Electronic copy available at: https://ssrn.com/abstract=3228827